JP2005043357A - Computed tomograph and computed tomography for classifying object - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To disclose a method and a system for detecting objects on the basis of computed-tomographic (CT) data. <P>SOLUTION: Sheet-like objects such as a sheet-like explosive can be detected by analyzing the density of a test voxel and an average density of voxels in its vicinity. The sheet-like objects can be also detected by erosion and expansion operation. Objects can be classified as to whether the objects are threats or not by computing corrected mass through the use of an average erosion density of the object and comparing it with a mass threshold value. Clump-like objects can by detected by a modified morphological Connected Component Labeling (CCL) approach for executing a series of erosion and expansion steps, individually labeling the objects, and separating data-adjacent objects so that analysis may be performed. The system can also distinguish objects containing liquids. The overall performance of the system including overall object detection rates and error alarm rates can be adjusted by individually adjusting the object detection rates and/or the error alarm rates. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

[Related applications]
This application is related to the following co-pending US application of the same assignee as the present application, the contents of which are hereby incorporated by reference in their entirety.

  Gregory L. "Nutting Slice CT Image Reconstruction Apparatus and Method" by Larson et al., US patent application Ser. No. 08 / 831,558 filed Apr. 9, 1997 (Attorney Docket No. ANA-118).

  Andrew P.M. “Computed Tomography Scanner System and Bearing” by Tybinkowski et al., US patent application Ser. No. 08 / 948,930, filed Oct. 10, 1997 (Attorney Docket No. ANA-128).

  David A. No. 08 / 948,937 (patent attorney docket number NA) filed Oct. 10, 1997, “Air Calibration Scan for Computed Tomography With Obstructing Objects” by Schaffer et al.

  Christopher C.I. "Computed Tomography Scanning Apparatus and Method With Temperature Comparison for Dark Current Offset" filed Oct. 10, 1997, US Patent No. 08/9489, filed by Ruth et al. .

  Christopher C.I. "Computed Tomography Scanning Target Detection Non-Parallel Slices" by Ruth et al., US patent application Ser.

  Christopher C.I. "Computed Tomography Scanning Target Detection Using Target Surface Normals" by Ruth et al., US patent application Ser.

  Christopher C.I. “Parallel Processing Architecture for Computed Tomography Scanning System Using Non-Parallel Slices” by Ruth et al., US patent application Ser. No. 08 / 948,697, filed Oct. 10, 1997 .

  Christopher C.I. US Patent Application No. 8 / National Patent Application No. 8 / National Patent Application No. 8 / November No. 8 of Oct. 10, 1997 135).

  Bernard M.M. "Computed Tomography Scanning Apparatus and Method Using Adaptive Construction Window" by Gordon et al., US patent application Ser.

  David A. “Area Detector Array for Computed Tomography Scanning System” by Schaffer et al., US patent application Ser. No. 08 / 948,450 filed Oct. 10, 1997 (Attorney Docket No. ANA-137).

  “Closed Loop Air Conditioning System for a Computed Tomography Scanner”, US patent application Ser. No. 08 / 948,692 filed Oct. 10, 1997 (patent attorney docket number NA-1) according to the invention of Eric Bailey et al.

  Geoffrey A. Legal et al., “Measurement and Control System for Controlling System Functions as a Function of Rotational Parameters, Patent No. 9th, Patent No. 9th, 1997, No. 10” ANA-139).

  Andrew P.M. “Rotary Energy Shield for Computed Tomography Scanner” by Tybinkowski et al., US patent application Ser. No. 08 / 948,698 filed Oct. 10, 1997 (Attorney Docket No. ANA-144).

  This application also relates to the following pending US patent applications, all filed on the same day as this application, assigned to the same assignee as the present application, and incorporated herein by reference in its entirety:

  “Apparatus and Methods for Eroding Objects in Computed Tomography Data” (Attorney Docket No. ANA-150) according to Sergey Simonovsky et al.

  Ibrahim M.M. “Apparatus and Methods for Combining Related Objects in Computed Tomography Data” (patent attorney docket number ANA-153) according to the invention of Bechwati et al.

  “Apparatus and Methods for Detecting Sheet Objects in Computed Tomography Data” (Act No. ANA-151) according to Sergey Simonovsky et al.

  Ibrahim M.M. “Apparatus and Methods for Classifying Objects in Computed Tomography Data Dependent Dependency Mass Threshold” No. 54, Patent Attorney No. 54 (Area No. 1).

  Ibrahim M.M. “Apparatus and Method for Correcting Object Density in Computed Tomography Data” (No. ANA-152, patent attorney number) according to the invention of Bechati et al.

  “Apparatus and Method for Density of Discrimination of Objects in Computed Tomography Data-Amplifying Density NA” by Sergey Simonovsky et al.

  “Apparatus and Methods for Detection of Computed Tomography Data” (patent attorney docket number ANA-148) according to the invention of Muzaffer Hiraoglu et al.

  "Apparatus and Methods for Optimizing Detection of Objects in Computed Tomography Data" (patent attorney docket number ANA-147) according to the invention of Muffer Hiraoglu et al.

  "Multiple-Stage Apparatus and Method for Detection Objects in Computed Tomography Data" (patent attorney docket number ANA-146) according to the invention of Muzaffer Hiraoglu et al.

  “Computed Tomography Apparatus and Method for Classifying Objects” by Sergey Simonovsky et al. (Patent attorney docket number ANA-155).

  “Appratus and Methods for Detecting Objects in Computed Tomography Data Using Erosion and Dilation of Objects” according to Sergey Simonovsky et al.

[Field of the Invention]
The present invention relates generally to computed tomography (CT) scanners, and more particularly to a target detection apparatus and method for a baggage scanning system using CT technology.

[Background of the invention]
Various x-ray baggage scanning systems are known for detecting the presence of explosives and other prohibited items in a baggage or travel baggage prior to loading the baggage on a commercial airplane. Many explosive materials can be characterized by a density range that is distinguishable from other items normally found in baggage, and explosives are generally easy to detect by X-ray equipment. A general technique for measuring the density of a substance is that when the substance is exposed to X-rays and the amount of radiation absorbed by the substance is measured, the absorbed amount indicates the density.

  Most of the X-ray baggage scanning systems in use today are of the “line scanner” type, where the source and detector array are stationary as the stationary X-ray source, stationary linear detector array, and baggage pass through the scanner. Includes a conveyor belt for transferring baggage between them. The x-ray source generates an x-ray beam that passes through the baggage and is partially attenuated by the baggage and then received by the detector array. During each measurement interval, the detector array generates data representing the integral of the density of the plane section of the baggage through which the X-ray beam passes, and these data form one or more raster lines of the two-dimensional image. Used to. As the conveyor belt transports the baggage past the stationary source and detection array, the scanner generates a two-dimensional image representing the density of the baggage viewed from the stationary detector array. The density image is usually displayed for analysis by a human operator or can be analyzed by a computer. Therefore, a very careful operator may be required to detect suspicious baggage. This careful attention increases operator fatigue, and as a result of fatigue and distraction, suspicious bags may pass through the system without being detected.

  Techniques using dual energy x-ray sources are known to provide additional information regarding the chemical characteristics of materials as well as density measurements. Techniques that use dual energy x-ray sources measure the x-ray absorption characteristics of a material for two different energy levels of x-rays. These measurements provide an indication of the atomic number of the material in addition to an indication of the density of the material. Dual energy X-ray techniques for energy selective reconstruction of X-ray CT images are described, for example, by Alvarez et al., “Energy-selective Reconstructions in X-ray Computerized Tomography” (Phys. Med. Biol. 1976, Vol. 5,733-744) and US Pat. No. 5,132,998.

  One proposed use for such dual energy technology is in connection with a baggage scanner that detects the presence of explosives in the baggage. Explosive materials are generally characterized by a known atomic number range and are therefore easy to detect with such dual energy x-ray sources. One such dual energy source is described in co-pending US patent application Ser. No. 08 / 671,202 (Attorney Docket No. ANA-094) entitled “Improved Dual Energy Power Supply”. Assigned to the same assignee as the present invention and incorporated herein by reference in its entirety.

  Certain types of explosives are particularly challenging for baggage scanning systems. This is because it can be formed into a geometric shape that is difficult to detect because of its moldable nature. Many explosives that can cause significant damage to an airplane are sufficiently large in length, width and height so that the X-ray scanner system can easily detect them regardless of the direction of the explosives in the baggage. Another problem with certain explosives is that they can be hidden in objects such as pieces of electronic equipment such as laptop computers. This may be difficult to detect with conventional line scanning techniques. Also, explosives that are powerful enough to damage an airplane can be formed into relatively thin sheets that are very small in one dimension and relatively large in the other two dimensions. In the image, especially when the material is arranged so that the thin sheet is perpendicular to the X-ray beam when the sheet passes through the system, it is difficult to see the explosive material, so it is difficult to detect the explosive part Sometimes.

  Systems that use CT technology typically include a third generation type of CT scanner, which typically includes an X-ray source and X-ray detector system fixed diagonally opposite an annular shaped platform or disk. Including. The disc is mounted in a rotatable manner in the gantry support, so that in operation, the disc is centered about the axis of rotation while X-rays pass from the source through the object located at the disc opening to the detector system Continue to rotate.

  The detector system can include a linear array of detectors arranged as a single row in the form of a circular arc whose center of curvature is at the focal point of the x-ray source, ie, a point within the x-ray source that emits x-rays. . The x-ray source emits from a focal point, passes through a planar projection field, and produces an x-ray fan-shaped beam that is received by the detector, ie, a fan beam. A CT scanner includes a coordinate system defined by X, Y, and Z axes that intersect at the center of rotation of the disk and are all perpendicular to each other as the disk rotates about the axis of rotation. This center of rotation is usually referred to as the “isocenter”. The Z axis is defined by the axis of rotation and the X and Y axes are defined by and within the planar projection field. Thus, a fan beam is defined as the volume of space defined between the point source, i.e. the focal point, and the light receiving surface of the detector of the detector array exposed to the x-ray beam. Since the dimension of the light receiving surface of the linear array of detectors is relatively small in the Z-axis direction, the fan beam is relatively thin in that direction. Each detector generates an output signal representative of the density of x-rays incident on that detector. Since the x-rays are partially attenuated by the total mass in the path, the output signal produced by each detector is the total mass located in the projection field located between the x-ray source and the detector. Represents density.

  As the disk rotates, the detector array is periodically sampled and each detector of the detector array produces an output signal representing the density of the portion of interest scanned during the interval. A collection of all output signals generated by all detectors in a detector array in a row at an arbitrary measurement interval is called “projection” and is the angular direction of the disk (and the corresponding X-ray source) And the angular direction of the detector array) is called the “projection angle”. At each projection angle, the X-ray path from the focal point, called a “line”, to each detector increases in cross section from the point source to the light receiving surface area of the detector, so the light receiving surface area of the detector area passes through the line. Because it is larger than the arbitrary cross-sectional area of the object to be measured, it is considered that the density measurement value is expanded.

  As the disc rotates about the object being scanned, the scanner produces multiple projected images at corresponding multiple projection angles. A well-known algorithm can be used to generate a CT image of interest from all projection data collected at each projection angle. A CT image represents the density of a two-dimensional “slice” of an object through which the fan beam has passed while the disk is rotated through various projection angles. The resolution of the CT image is determined, in part, by the width of the light receiving surface area of each detector in the plane of the fan beam, the width of the detector being the dimension measured herein in the same direction as the width of the fan beam. The length of the detector is defined herein as the dimension measured in a direction perpendicular to the fan beam and parallel to the rotation of the scanner or the Z axis.

  Baggage scanners using CT technology have been proposed. One approach described in US Pat. Nos. 5,182,764 (Peschmann et al.) And 5,367,552 (Peschmann et al.) (Hereinafter referred to as the '764 and' 552 patents) is: Developed commercially, hereinafter referred to as “invision machine”. The invision machine includes a third generation type CT scanner, which includes an X-ray source and an X-ray detector system fixed diagonally opposite each of an annular shaped platform or disk. The disc is mounted in a rotatable manner in the gantry support, so that during operation, the disc is centered on the axis of rotation while X-rays pass from the source through the object located in the disc opening to the detector system. Continue to rotate

  One important design criterion for baggage scanners is the speed with which the scanner can scan items in the baggage. For practical use at any major airport, the baggage scanner must be able to scan a large number of bags at very high speeds. One problem with invision machines is that a CT scanner of the type described in the '764 and' 552 patents can generate data for one sliced CT image by rotating the disk one revolution, for example. It takes a relatively long time, such as about 0.6 to about 2.0 seconds. Further, the thinner the slice of the beam that passes through each image bag, the better the resolution of the image. CT scanners must provide images with sufficient resolution to detect plastic explosives on the order of only a few millimeters thick. Therefore, many rotations are required to provide sufficient resolution. In order to meet the high throughput rate of baggage, conventional CT baggage scanners such as invision machines can only generate a few CT images per bag. Clearly, the entire bag cannot be scanned within the allotted time for fairly fast throughput. If only a few CT images are generated for each baggage item, the majority of the item will remain unscanned, so there will be sufficient scan to identify all potential dangerous goods in the bag, such as sheet-like explosives. Not provided.

  In order to improve throughput, the invision machine uses a preliminary screening process, which produces a two-dimensional projection image of the entire bag from one angle. The projection area identified as potentially containing dangerous objects can now be completely scanned or manually inspected. This pre-screening and selective area scanning approach does not scan the entire bag and therefore allows potential dangerous goods to pass through undetected. This is especially true when the sheet item is oriented to traverse the direction of propagation of the radiation used to form the preliminary screening projection and the sheet covers a relatively large portion of the bag area.

  Another baggage scanning system is an international patent application issued May 2, 1996 entitled "X-Ray Computed Tomography (CT) System for Detecting Thin Objects" by Eberhard et al. WO 96/13017 (hereinafter referred to as “Eberhard et al. System”). In the Eberhard et al. System, the entire bag is subjected to a CT scan to generate voxel density data for the bag. A connected component labeling (CCL) process is then applied to the entire bag to identify objects by grouping voxels physically close to each other and having a density within a predetermined density range. Next, the volume of each object is determined by counting the voxels of each object. If the volume of the object exceeds the threshold, the object mass is calculated by multiplying the volume of each object voxel by its density and then summing the individual voxel masses. If the subject's mass exceeds the mass threshold, the subject is concluded to be a threat.

  Eberhard et al. Publications teach that the system can identify thin objects. The system can set its labeling density to a low level and thus detect thin objects seen from the edge that partially fills the voxel.

  A significant drawback of the Eberhard et al. System is that it may lose sight of thin objects such as sheet explosives that do not see from the edges and cover a large area of the bag. Such laterally oriented sheet objects only slightly increase the density measured with the bag and have only a small density contrast with the background. If the density threshold used during CCL is set low enough to detect these sheets, the contrast between the sheets and the background is small, so the entire bag is connected to each other and labeled, Recognizable objects are not identified. If the threshold value is set higher than this, the sheet object is lost.

  It would be advantageous to automatically analyze the density data acquired by the baggage scanning device to determine whether the data indicates the presence of any import / export prohibited items such as explosives. This automatic explosive detection process must have a relatively high detection rate so that the likelihood of losing explosives in the bag is small. At the same time, the false alarm rate of the system must be relatively low to greatly reduce or eliminate false alarms on harmless items. Due to practical considerations on baggage throughput at large private airports, a high false alarm rate can reduce the performance speed of the system to a prohibitively low rate. It would also be advantageous to implement a system that can distinguish between different explosive types, such as powders, lumps, sheets, and so more accurately characterize detected threats.

[Summary of Invention]
The present invention is directed to an object identification apparatus and method, and a computer linked tomography (CT) baggage scanning system and method using the object identification apparatus and method of the present invention. The object identification apparatus and method of the present invention analyzes acquired CT density data for a region where a data object is detected. The region can include at least a portion of the inside of the container, such as a piece of baggage or travel luggage. The detected object can now be labeled according to its physical shape. For example, in one embodiment, a label can be affixed to the object as a bulk material or sheet material. In one embodiment, the objects are detected and labeled and then identified, that is, classified as a threat or non-threat object.

  In one embodiment, the present invention uses a sheet detection process that identifies thin sheet shaped objects. One form of sheet detection applies a static approach to determine whether each volume element or “voxel” of density data is associated with a sheet object. In this static approach, each voxel is analyzed by comparing its density with the density of neighboring voxels. In one embodiment, the average and standard deviation of the density of adjacent voxels is calculated. The difference between the density of the analyzed voxels and the average density of adjacent voxels can be compared with a predetermined threshold difference and correlated with the standard deviation of the density of adjacent voxels. If the difference between the density of the voxel in question and the average density exceeds a predetermined threshold difference, it is concluded that the voxel in question is associated with a thin object such as a sheet.

  Voxels can be analyzed one at a time and individually labeled depending on whether they are associated with a sheet object. Next, the set of labeled voxels can be analyzed and related voxels can be grouped into subjects. In one embodiment, a standard connected component labeling (CCL) approach is used to group adjacent voxels of similar density into a sheet. In this standard CCL approach, each voxel labeled as a sheet voxel is compared to the voxel of an adjacent sheet to determine the density difference. If the density difference is below a predetermined density difference threshold, it is estimated that two adjacent voxels belong to the same object, i.e. a sheet. This process is continued until all voxels labeled as sheet voxels are combined into a sheet object. As a result, one or more objects may be identified within one region or bag data.

  The apparatus and method of the present invention can also classify objects such as detected sheet objects as threat or non-threat objects. In one embodiment, this is done by comparing the mass of interest with a predetermined threshold mass. If the mass of the object exceeds a predetermined mass threshold, the object is concluded as a threat object. If the bag is identified as containing a threat object, it can be marked for further analysis. The bag can be identified for further inspection by the operator, or an image of the entire interior of the bag can be generated from the density data.

  The present invention also provides identification and classification of massive objects, such as massive explosives, in CT density data acquired for areas such as the interior of a piece of travel luggage or baggage. The mass detection process of the present invention uses a modified connected component labeling (CCL) process to identify mass objects. In standard CCL, adjacent voxels whose density is less than a predetermined threshold are labeled as part of the same object. Each voxel is analyzed and compared to its neighbors, voxels are combined and targeted. This general CCL approach has the disadvantage that objects that are close to or in contact with each other and that have similar densities may be combined into one object. The modified CCL approach of the present invention separates these objects into individual labeled objects.

  The approach of the present invention applies a “morphological” CCL method. Each subject first “erodes” by removing all of its surface voxels. This tends to separate the connected objects into a plurality of individual objects. Next, the separated objects are labeled separately. Next, an “expansion” step is applied that adds and returns surface voxels to the identified and labeled object. Thus, with this morphological approach to CCL, objects in close proximity to each other can be separately identified and labeled. Now we can recognize the objects separately and classify them as threatening or non-threatening.

  Standard erosion approaches used in other image data processing situations may produce undesirable results. For example, one standard erosion process identifies surface voxels as voxels having at least one adjacent voxel whose density is below a predetermined threshold. This assumes that all voxels classified as being in the vicinity of adjacent voxels below the threshold are surface voxels. These identified surface voxels are then removed from the object. The disadvantage of this approach is that there are situations where voxels that are not on the surface of the object are removed. For example, in the case of an object having a void area inside, such as a cylindrical object or a rod-shaped object having a thin cylindrical hole in the axial direction, voxels around the outside of the void area are removed. The undesirable result is that the internal void area is enlarged by the erosion process.

  In one aspect of the invention, erosion is performed in such a way that the likelihood of removing non-surface voxels is reduced. In this aspect of the invention, a plurality of adjacent voxels are identified for each voxel. In one embodiment, adjacent voxels define a three-dimensional auxiliary area or neighborhood surrounding the voxel in question. The auxiliary area may have a cubic shape. Each voxel in the auxiliary area is analyzed to determine if its density is within one or more predetermined ranges of density. For each voxel in question, the number of voxels in the associated auxiliary area whose density falls within a predetermined density range is compared to a threshold value. If the number is below the threshold, it is concluded that the voxel in question is the surface voxel of interest and the voxel is removed from the subject.

  In one embodiment of the erosion process of the present invention, the predetermined density range is determined based on the density of the voxels in question. The range is selected to be a range that includes the density of the voxels in question. In this case, the analysis then determines the number of voxels in the auxiliary area that are in the same density range. If the number does not exceed the threshold, it is concluded that the voxel in question is on the surface of the object and the voxel is removed from the object.

  In one embodiment of the erosion process of the present invention, the predetermined density range is selected from a plurality of ranges, each range being defined based on the threat to be identified. In this embodiment, the density of the voxels in question determines the potential threat material and thus the selected density range. For example, if the density of the voxel in question indicates that it is that of a massive explosive, the bulk explosive density range is selected for analysis of the auxiliary region surrounding the voxel in question. If the number of voxels in the auxiliary region that are within range and are therefore part of the same massive explosive object does not exceed the threshold, the voxel in question is concluded as a surface voxel and is removed from the object. This erosion approach according to the present invention reduces the possibility of the subject's internal voids being enlarged and increases the possibility of removing only external surface voxels.

  In another aspect of the invention, the expansion step of the morphological CCL approach is applied to more accurately measure the size of the object and hence its mass. In this approach, when the voxel is added back to the surface of the eroded object, the density designated as an additional voxel is the average erosion density of the bulk object. That is, the average density of all voxels of the eroded object is calculated. During subsequent expansion, each voxel added to the surface of the eroded object is assumed to have a density of average erosion density. This approach greatly reduces or eliminates object mass and density inaccuracies caused by partial volume effects caused by surface voxels that average the density of the object and background contained within one surface voxel.

  In another aspect of the invention, sheet-like objects can be detected with density data using a morphological approach similar to the morphological CCL applied to the detection of massive objects. In this morphological sheet detection approach, all objects in the data are eroded a predetermined number of times so that thin sheet objects are removed from the data. The number of erosion performed is based on the number of erosion required to remove the sheet material from the data, which is related to the thickness of the sheet. Each erosion can remove one layer of surface voxels. Thus, the number of erosion is related to the expected sheet thickness and voxel size. After performing the full erosion step, the voxels remaining in the data are assumed to be associated with the massive object. Next, expansion can be performed to restore the massive object to its original size. The data associated with these objects can now be deleted from further processing. Remove the bulk object from the original data, then analyze and label the sheet object. The remaining voxels are analyzed one at a time, similar to the CCL process, and the voxels are combined into a sheet object, and then the sheet object is labeled. Next, the sheet object is distinguished and classified as threatening or non-threatening, for example, by comparing the mass of the object with a predetermined mass threshold. A sheet with a mass above a threshold can be classified as a threat.

  An optional CCL step can be performed between the erosion step and the expansion step to identify objects with the eroded data. The subsequent expansion and subtraction steps may then be performed only on objects that exceed a predetermined size or mass.

  Thus, according to the present invention, at least two sheet detection processes can be applied to region data to identify voxels associated with sheet-shaped objects. The two approaches include the CFAR method and the morphological erosion expansion method described above. Either approach can generate a set of binary data associated with the voxels, which can define each voxel as part of a sheet or part of a sheet. After identifying the voxels of the sheet, a voxel connection approach, such as morphological CCL, standard CCL, or other connectivity method of the present invention, is performed to connect and target the voxels. This approach identifies the sheet in the data before combining the voxels into the object, so the target connection process does not remove the sheet from the data and does not make detection impossible. Note that the connection approach can be applied to binary data generated by the sheet detection method, or to the product of binary data and density data, i.e., density data of voxels identified as sheet voxels.

  In another aspect of the invention, separate objects that should be considered as a single threat are combined or merged. A particular threat includes a plurality of targets, such as a plurality of rod-shaped targets that are bundled or otherwise coupled together. These objects can be separated from each other during the erosion step of the morphological CCL process, so that they are considered as separate objects, and each is not categorized as a threat alone in a mass threshold manner. However, combining these targets is a threat and must be classified as such. The merge process of the present invention combines such distinct objects to identify and identify them as threats.

  In one embodiment, the merge process of the present invention identifies objects that are close to each other and have similar or equal density and combine them into one object. In one embodiment, a bounding box is calculated for each object. Compare subjects for similar density. If the difference in object density is below a predetermined threshold and the absolute density of one or both objects is within a predetermined density range that defines the threats of multiple objects, the distance between the bounding boxes is determined. If the distance between bounding boxes is below a predetermined threshold and the object is considered sufficiently close to be considered a single object, it is concluded that the objects should be combined into a single object. Calculate the total mass of all individual objects and compare to the threat mass threshold. If the total mass is above the threshold, it is concluded that the combined target is a threat.

  In another aspect, the present invention can merge multiple small sheet objects into a single sheet object. According to the present invention, unlike the prior art partial or two-dimensional analysis, the analysis of a three-dimensional CT image of an actual bag can obscure or obstruct the image of a large dense object such as a metal bar. In other words, the effect of being represented as multiple images of separate individual sheets is identified. As a result, one large sheet-like object may be identified as a plurality of smaller objects. Multiple small objects may be small enough, i.e., sufficiently small in mass, and therefore all classified as non-threat items. This is particularly a problem when an object should be classified as a threat, and will be classified as such if the system recognizes it as one object rather than multiple separate objects. In order to solve this problem, in one aspect of the present invention, each sheet-like object is associated with a surface. When multiple sheets are detected in the data, the surface of each sheet is inspected in a three-dimensional space. If the planes intersect and the intersection is close to the sheet, it is concluded that the individual sheet is actually part of a larger sheet. The masses of the individual sheets are combined into a single value and compared to the mass threshold during identification. When the combined sheet mass exceeds the mass threshold, it is concluded that the sheet is a threat.

  As described above, after identifying an object with density data, it is classified whether it should be regarded as a threat. Generally, mass discrimination is used to classify objects. In one embodiment, the mass of each identified object is calculated by multiplying the density of each voxel by its volume and then summing the entire individual voxel mass. Next, the total mass of the object is compared to a mass threshold. If the mass of the object is above the threshold, it is concluded that it is a threat object.

  In the present invention, the mass threshold used for an object can be determined based on the type of object. That is, different mass thresholds are used depending on the type of object. For example, a sheet-like object can be compared to one threshold and a powdered explosive can be compared to a different mass threshold. This is because different explosives pose different threats depending on their mass. A large amount of one type of explosive may not pose a more serious threat than a different type of less explosive. Thus, in the present invention, the mass threshold can be selected based on the type of explosive. In one embodiment, the selection of the mass threshold is determined by the density of the identified object. This is because density is closely related to the type of object being identified. That is, the density of one type of explosive is generally different from the density of another type of explosive. These individual densities are used to identify the type of explosive and thus determine the mass threshold to be used in classifying the subject as a threat. This mass threshold setting, which depends on the density of the present invention, provides a much more accurate threat classification than previous systems that used one mass threshold for all subjects.

  In another aspect of the invention, the calculation of the total mass of the object is enhanced and the threat classification accuracy of the system is improved. As described above, the modified CCL of the present invention described herein can be used to erode surface voxels of an object from the object. According to the present invention, an erosion step can be performed to eliminate the effect of partial volume voxels located on the surface of the object. These voxels introduce inaccuracies. This is because the density value includes density components from both the object and the background at the object boundary. In this aspect of the invention, erosion is performed to remove surface voxels. Next, the average erosion density of the remaining target voxels is calculated. The average erosion density is the average density of voxels remaining in the subject after the erosion step. The eroded surface voxels are then replaced with voxels having a density value equal to the average erosion density. Next, the total mass of the object is calculated using surface voxels with average erosion density values. This corrected total object mass provides a more accurate classification of the object during subsequent mass identification.

  In another aspect of the invention, the separation between multiple objects in close proximity to each other is improved during the CCL process by carefully adjusting the acceptable density range of the object voxels. In one embodiment, the allowed density is defined within a plurality of density ranges, with the density range having gaps between densities that are not allowed to associate the subject matter with the voxel. That is, voxels with a density in the gap are rejected, and voxels in one of the density ranges are allowed as belonging to the subject in question. The allowed density range can be selected according to the density of known threat objects. For example, a density range can be selected for each of several different types of known explosives. In one embodiment, the gap between density ranges is selected to match the density expected for a typical surface voxel. By rejecting these surface voxels, adjacent objects that would otherwise be combined and labeled as one object can be separated and labeled as individual objects. This can be analyzed independently as a separate subject and classified according to the threat level it presents. Thus, more accurate classification can be achieved using multiple density ranges.

  In yet another aspect of the invention, the liquid material may not pose a threat and may be classified as a non-threat object. Thus, the subject identification and classification system of the present invention can recognize and identify liquids in containers so that they can be deleted as a threat. This provides a way to distinguish detected objects beyond the mass and density discrimination approach of the present invention described above.

  In one embodiment, the present invention determines whether the object is a contained liquid by first generating a bounding box that surrounds the object. Calculate the number of voxels near each surface of the bounding box. Next, the top surface of the liquid is identified by identifying the horizontal surface of the bounding box. The ratio of voxels near the top surface to the total number of surface voxels can now be calculated. If the top voxel fraction exceeds a predetermined threshold ratio, and if the density of voxels above the top surface indicates that there is air above the top surface, the subject is concluded to be contained liquid. In one embodiment, this can conclude that the subject is not a threat.

  In another embodiment, the present invention applies a statistical approach to determine if the object in the bounding box is an contained liquid. A histogram for the top and bottom voxels is calculated along the line between the top and bottom of the bounding box. The peak of the top histogram indicates the vertical position of the top surface voxel, and the peak of the bottom surface histogram indicates the vertical position of the bottom surface voxel. If the ratio of the number of top voxels to the top surface area in the bounding box exceeds a threshold and the ratio of the number of top voxels to the bottom voxel exceeds another threshold, then conclude that the subject is an contained liquid it can.

  In yet another aspect of the invention, detection is performed in multiple paths or stages so that the overall detection process is more efficient. Each item that can be identified by the method of the present invention is generally associated with a unique set of detection steps. In a typical detection approach, all acquired CT density data is applied sequentially to each detection approach. This is understood to be very time consuming. Also, if an object is analyzed by one particular approach and classified according to the threat it presents or its object type, reanalyzing the set of data with the other remaining detection processes becomes inefficient. In the present invention, a multipath or multi-stage detection approach is used to eliminate these inefficiencies. The specific detection steps used to identify specific items are applied separately and in one embodiment are applied in parallel. In one particular embodiment of this multipath method of the present invention, where one particular detection path is applied to a set of data and a portion of the data is classified, the classified portion of the data is removed from further processing. Remove. This eliminates the inefficiency caused by unnecessary reanalysis of already classified data.

  The present invention can also optimize the overall system detection rate (detection probability) and false alarm rate. Each item that can be detected by the system of the present invention is associated with an individual detection rate and false alarm rate. For example, detection of sheet explosives has a unique detection probability and false alarm rate. Also, each individual explosive type has its own detection probability and false alarm rate. The overall system detection probability is the accumulation of each individual detection rate, and in one embodiment is the average of the individual detection rates. Also, the overall false alarm rate of the system is the accumulation of all individual false alarm rates, and in one embodiment this is the sum of the individual false alarm rates. In the present invention, the overall detection rate can be optimized by adjusting one or more of the individual detection rates. Also, the overall false alarm rate can be optimized by adjusting one or more of the individual false alarm rates. Thus, overall system performance can be adjusted as needed, and individual detection rates and / or false alarm rates can be adjusted to achieve a desired overall detection rate and / or false alarm rate. it can.

  One or more individual detection rates can be lowered below a specified overall detection rate. The system can provide the flexibility to adjust one or more individual detection rates to lower levels while maintaining the overall rate within specified limits. Decreasing one detection rate can also reduce the associated false alarm rate. Thus, the overall system false alarm rate can be reduced while maintaining the overall system detection rate within specified limits. Also, the overall detection rate can be maintained at a particular value while adjusting the individual and / or overall system false alarm rate to a desired level.

  In addition to the advantages described above, the present invention provides substantial advantages over the preceding systems described above. For example, the system of the present invention provides a complete CT scan of the bag and can therefore analyze the complete 3D image data of the bag. As a result, the system can detect objects such as thin sheets in a bag, regardless of direction and size. In the invision machine, only a region identified as suspicious in the two-dimensional preliminary screening is subjected to a three-dimensional scan. Also, in one embodiment of the present invention, voxels are not connected and are identified as multiple objects until a voxel belonging to a thin sheet object is first identified. This eliminates sheet identification problems found in systems such as Eberhard et al.

  The above and other objects, features and advantages of the present invention will become apparent from the following more detailed description of the preferred embodiments of the present invention, illustrated in the accompanying drawings. In the drawings, like reference characters refer to the same parts throughout the different views. The drawings are not necessarily to scale, emphasizing the principles of the invention.

Detailed Description of Preferred Embodiments of the Invention
The present invention provides an apparatus and method for detecting, identifying and / or classifying objects in a region of CT data. The region can include the interior of a piece of baggage or travel baggage that is transported to or inspected to a commercial aircraft. Thus, the present invention can be implemented with a CT baggage scanning system. Objects identified by the present invention are those that are known to pose a threat to people at the airport or on board. This object can include explosive objects and materials.

  Note that the explosive objects and materials that can be detected by the present invention can be of various shapes and materials. Explosives may be commercially produced, military, or improvised, ie homemade. For example, explosive objects may be in various shapes, such as sheets, single cylindrical containers or other such shapes, multiple cylindrical or other rod shapes, and other massive shapes, It is not limited to these. Various types of explosive substances formed in or included in such a shape can be detected by the present invention.

  Throughout the following description, it can be seen that the various methods of the present invention are implemented using many threshold values, such as density threshold, mass threshold, density dependent mass threshold, and difference threshold, as well as process parameters. These thresholds and parameters are determined based on a comprehensive analysis of CT data, such as actual 3D CT density data, for many actual threatening and non-threatening objects. This analysis includes statistical analysis of the data using statistical methods such as simulated annealing and general algorithms. In accordance with the present invention, this analysis allows thresholds and / or parameters based on the specific purpose to be met, eg, false alarm and / or detection rate setting / optimization, distinction of explosive types, as described below. Can be selected.

  1, 2 and 3 include a perspective view, an end cross-sectional view and a radial cross-sectional view, respectively, of a baggage scanning system 100 constructed in accordance with the present invention, which provides object detection, identification and classification according to the present invention. To do. Baggage scanning system 100 generates CT data for an area that can include a piece of baggage. The system can use the CT data to generate an image volume element or “voxel” of the region. Baggage scanning systems are described above and incorporated by reference in co-pending US patent application Ser. Nos. 08 / 831,558, 08 / 948,930, 08 / 948,937, 08 / 948,928, 08 / 948,491, 08 / 948,929, 08 / 948,697, 08 / 948,492, 08 / 949,127, 08 / 948,450, 08 / 948,692, 08 / 948,493, and 08 / 948,698.

  System 100 includes a conveyor system 110 that continuously transports baggage or travel items 112 in the direction indicated by arrow 114 through the central port of CT scanning system 120. The conveyor system includes a motor driven belt that supports baggage. Although the conveyor system 110 is illustrated as including a plurality of individual conveyor sections 122, other forms of conveyor systems may be used.

  The CT scanning system 120 rotates about an axis of rotation 127 (shown in FIG. 3), preferably parallel to the direction of travel 114 of the baggage 112, so an annular shaped turntable or disk 124 disposed within the gantry support 125. including. The disc 124 may be issued to any suitable drive mechanism, such as belt 116 and motor drive system 118, or Gilbert McKenna on Dec. 5, 1995, assigned to the assignee of the present application and as a whole by reference. Other suitable drive mechanisms, such as those described in US Pat. No. 5,473,657 (Attorney Docket No. ANA-30CON) entitled “X-ray Tomographic Scanning System” incorporated herein Thus, the rotary shaft 127 is driven around. The turntable 124 defines a central port 126 through which the conveyor system 110 transports the baggage 112.

  The system 120 includes an x-ray tube 128 and a detector array 130 disposed diagonally opposite the platform 124. Detector array 130 was filed on Oct. 10, 1997 and is co-pending US patent application Ser. No. 08 / 948,450 entitled “Area Detector Array for Computed Tomography Scanning System” (Attorney Docket No. ANA-137). A two-dimensional array such as the array described in The system 120 further receives a CT data signal generated by the detector array 130 and processes a data acquisition system (DAS) 134 and an x-ray control system that provides power to the x-ray tube 128 and controls its operation. 136 is included. The system 120 is also preferably provided with a computer processing system that processes the output of the data collection system 134 and generates the necessary signals for operation and control of the system 120. The computer system can also include a monitor that displays information, such as generated images. X-ray tube control system 136 is assigned to the same assignee as the present patent application and is co-pending US patent application Ser. No. 08/671, entitled “Improved Dual Energy Power Supply”, which is incorporated herein by reference in its entirety. It may be a dual energy X-ray tube control system, such as the dual energy X-ray tube control system described in 202 (patent attorney docket number ANA-094). While dual energy X-ray technology that selectively reconstructs X-ray CT images is particularly useful for indicating the atomic number of a material in addition to indicating the density of the material, the present invention provides this type of control. It shall not be restricted to the system. It can be seen that the detailed description herein of the object identification and classification system and method of the present invention provides details in connection with single energy data. It is understood that the description is applicable to multi-energy technologies. The system 120 also includes a shield 138, which may be manufactured from lead, for example, to prevent radiation from propagating beyond the gantry 125.

  In one embodiment, the x-ray tube 128 produces an x-ray pyramid beam 132, often referred to as a “conical beam”, in which the baggage 112 is transported by the transport system 110 in three dimensions. Pass through the imaging field. The cone beam 132 is received by the detector array 130 after passing through the baggage located in the imaging field, which generates a signal representative of the density of the exposed portion of the baggage 112. Thus, the beam defines a scanning volume of space. The platform 124 rotates about its axis of rotation 127, thereby generating multiple projections at corresponding multiple projection angles as baggage is continuously transported through the central port 126 by the transport system 110. The X-ray source 128 and the detector array 130 are transported in a circular orbit around the baggage 112.

  In a well-known manner, signals from detector array 130 can first be collected by data acquisition system 134 and then processed by a computerized processing system using CT scanning signal processing techniques. . The processed data can be displayed on a monitor or further analyzed with a processing system as described in detail below to determine the presence of a suspicious substance. For example, CT data can be analyzed to determine if the data suggests the presence of a substance having an explosive density (and molecular weight if using a dual energy system). If such data is present, an automatic ejector (not shown) that removes the suspicious bag from the conveyor, for example, sounds an audible alarm or provides a visual alarm display on the monitor screen, or for further inspection. Appropriate means can be provided to indicate the detection of such material to an operator or system monitor by providing or stopping the conveyor to inspect and / or remove suspicious bags.

  As described above, the detector array 130 may be a two-dimensional detector array that can provide scan data in both the X and Y axis directions, as well as in the Z axis direction. At each measurement interval, a plurality of detector array 130 columns generate data from the corresponding plurality of projections while simultaneously scanning the volumetric region of baggage 112. The size and number of detector rows are preferably selected as a function of the desired resolution and throughput of the scanner, which is a function of the rotational speed of the turntable 124 and the speed of the transport system 110. These parameters are the time required for the table 124 to make one complete rotation, and the volumetric area scanned by the detector array 130 during one table rotation will cause the detector array 130 during the next rotation of the table. Preferably, the transport system 110 is selected to advance the baggage 112 sufficiently so that it is adjacent and non-overlapping (or partially overlapping) with the volumetric area to be scanned.

  The transport system 110 continuously transports the baggage items 112 through the CT scanning system 120, preferably at a constant speed, while the platform 124 rotates continuously around the baggage items that pass through, preferably at a constant speed. In this way, the system 120 performs a helical volumetric CT scan of the entire baggage item. Baggage scanning assembly 100 preferably generates a volumetric CT display of the entire baggage item using at least a portion of the data provided by array 130 and a helical reconstruction algorithm as the baggage passes through the system. In one embodiment, system 100 is a co-pending US patent filed Apr. 10, 1997, entitled “Nutting Slice CT Image Reconstruction Apparatus and Method” and incorporated herein by reference. Perform nutritive slice reconstruction (NSR) on the data as described in application Ser. No. 08 / 831,558 (Attorney Docket No. ANA-118). Thus, the system 100 provides a full CT scan of each bag, rather than providing only a CT scan of a selected portion of the baggage item, without the need for a preliminary screening device. The system 100 also provides rapid scanning because the two-dimensional detector array 130 allows the system 100 to scan a relatively large portion of each baggage item simultaneously with each turn of the platform 124.

  FIG. 4 includes a mechanical / electrical block diagram of one embodiment of the baggage scanning system 100 of the present invention. The mechanical gantry of the scanner 100 includes two main components: a disk 124 and a frame (not shown). The disk 24 is a rotating element that carries an x-ray assembly, a detector assembly 130, a data acquisition system (DAS) 134, a portion of a high voltage power supply and monitor / control assembly, a power supply assembly and a data link assembly. The frame supports the entire system 100 including the baggage handling conveyor system 110. The disk 124 is mechanically connected to the frame via a double angle contact ball bearing cartridge. The disk 124 can be rotated at a constant speed by a belt that can be driven by a DC servo motor 505. The gantry also includes an x-ray shield on the disk and frame assembly.

  In one embodiment, the baggage conveyor system 110 includes a single belt driven at a constant speed to meet specified throughput requirements. The belt can be driven by a high torque, low speed assembly to provide a constant speed under varying load conditions. Low attenuation carbon graphite epoxy material can be used in the X-ray part of the conveyor bed. The overall length of the conveyor is designed to accommodate three bags of average length. Use tunnels around the conveyor to meet the appropriate safety requirements of the cabinet-type X-ray system.

  In one embodiment, 208 volt, three phase, 30 amp input power serves as the main power source capable of supplying power to the entire system. This input power can be provided by the airport where the system is installed. Power is transferred from the frame through a series of frame brushes that are in continuous contact with a metal ring attached to the disk 124. A low voltage power source 501 on the disk 124 provides power to the DAS 134, the X-ray cooling system and various monitor / control computers and electronics. A low voltage power supply on the frame provides power to the reconstruction computer and various monitor / control electronics. The conveyor motor 503, gantry motor 505, high voltage power supply and X-ray coolant pump can all supply power directly from the main power supply.

  The high voltage power supply provides power to the X-ray tube 128. The power supply can provide a double voltage between the cathode / anode. The drive waveform may be any desired shape, preferably a sinusoidal shape. This power supply can also provide x-ray filament power. The power supply current can be kept approximately constant at both voltages.

  Dual energy x-rays strike the baggage and some of the x-rays pass through and strike the detector assembly 130. The detector assembly 130 performs analog conversion from X-rays to visible photons and further to current. The DAS 134 can sample the detector current, multiplex the amplified voltage into a set of 16-bit analog-to-digital converters, and multiplex the digital output to a computerized processing system 515, which can be: In accordance with the present invention, CT data is generated and processed to detect, identify and classify objects within a piece of baggage 112 as described in FIG. In one embodiment, digital data from DAS 134 is transferred to processing system 515 via contactless serial data link 511. The DAS 134 can be triggered by the angular position of the disk 124.

  Contactless links 511 and 513 can transfer high speed digital DAS data to the processing system 515 and low speed monitor / control signals between the disk and frame control computers. Data link 511 may be based on an RF transmitter and receiver.

  In one embodiment, the image reconstruction portion of processing system 515 converts the digital line integral from DAS 134 into a set of two-dimensional images of bag slices, both at high energy and low energy. CT reconstruction can be performed via helical cone beam decomposition, such as the nutritive slice reconstruction method described in co-pending US patent application Ser. No. 08 / 831,558, incorporated by reference above. . The reconstructor can include embedded software, a high speed DAS port, an array processor, a DSP based convolver, an ASIC based rear projector, an image memory, a UART control port, and a SCSI output port for image data. The array processor can perform data correction and interpolation. The reconstructor is self-hosted and can tag images based on baggage information received via the UART interface with the frame computer.

  The processing system 515 can include a PC-based embedded control system. All subsystems can monitor important health and status information. The system can also control both operating systems, sense baggage information, control environments such as temperature and humidity, sense the angular position of the disk 124, and trigger DAS and HVPS. The system can also have a video and keyboard interface for engineering diagnostics and control. A control board can also be included for field service.

  Most types of explosives can be grouped into several categories based on their shape and / or constituent materials. For example, categories can be based on shape, including sheet, bar, chunk, and other categories. Certain types of materials can be further divided into subtypes, which are also based on containers such as cylinders. These categories have a variety of typical characteristics such as shape, size, mass or density. In general, one detection approach, such as the previous approach described above, cannot efficiently detect all of these explosive types. In one embodiment, the present invention includes multiple separate detection paths, which can include separate paths for each type. For example, the method can include a sheet explosive path, and a remaining explosive path, referred to herein as a “bulk”.

  In one embodiment of the object detection method and apparatus of the present invention, the process first performs partial differentiation of the data to identify sheet-shaped objects. Next, a connection step such as a CCL format is executed to connect the objects. Then further differentiation is performed to classify the identified objects according to the potential threat. This is in contrast to previous systems, such as Eberhard et al., Which first performs a connection and then performs a distinction, thereby losing a thin sheet object.

  The basic steps of the method according to one aspect of the invention include sheet explosive detection, bulk explosive detection, and differentiation. In one embodiment, sheet detection and mass detection can be performed separately along two parallel paths. In one embodiment, the detection of sheet explosives is based on a process known as constant false alarm rate method (CFAR) and is modified by the present invention to statistically determine whether a volume element or voxel belongs to a sheet explosive. To decide. Sheet voxels can also be identified with the morphological sheet detection approach according to the present invention described in detail below. In one embodiment, the voxels identified as sheet voxels in CFAR or morphological sheet detection of the present invention are then connected and labeled using a standard connected component labeling (CCL) process. . In another embodiment, the morphological CCL of the invention described herein is used to connect and label the voxels. The labeled object can then be distinguished by its mass. If the mass of the object is greater than a predetermined threshold, the object is declared a sheet explosive.

  In one embodiment of the invention, the detection of massive explosives uses a modified connected component labeling (CCL) process that includes morphological scanning (erosion and expansion) to prevent objects from growing together. In one embodiment, multiple sheets are detected in separate analysis paths and therefore do not need to be held in erosion and expansion steps. Mass detection can also involve object merging under the control of spaced and threatening objects, for example, individual bar objects to be considered as one object. The distinction is based on the density and mass of the detected object. In one embodiment, the mass threshold for discrimination depends on density. For several reasons, relatively low density objects can be assigned a higher mass threshold. For example, the data indicates that the amount of low density explosives required to produce a particular amount of damage is greater than the amount of high density explosives. Thus, at lower densities, a greater amount of material is required to trigger an alarm condition. In addition, the false alarm rate may be high at low density. Therefore, increasing the mass threshold can reduce the number of false alarms at low density.

  FIG. 5 includes a top level flow diagram illustrating the logic flow of one embodiment of the subject identification method of the present invention. In one embodiment, in a first step 301, the reconstructed CT image data is received and analyzed to define a region of interest (ROI) or region bounding box. This process removes voxels outside the bag, thus greatly reducing the size of the data set. The method can then proceed along parallel paths including a sheet-like object detection path and a block-like object detection path.

  A sheet-shaped object is detected in the sheet detection step 302 along the sheet detection path. In the discrimination step 306, the detected object is analyzed to determine whether it is a threat. In one embodiment, this is done by comparing the mass of interest to a mass threshold. The distinguishing step 306 generates bag label image data, which marks the voxels belonging to each sheet-like object and identifies the physical properties (preferably density and mass) of each sheet-like object and its position in the bag. Identify. The label image data for each voxel also includes a number that identifies the voxel according to the identified object or identifies the voxel as background.

  In the lump detection step 304, a lump type object is detected along the lump detection path. Next, in the distinguishing step 308, the detected massive object is analyzed to determine whether it is a threat. The distinguishing step 308 generates bag label image data, which marks the voxels belonging to each mass object and identifies the physical properties (preferably density and mass) of each mass object and its location within the bag. .

  The decision data fusion step 310 of the method obtains the label image data generated in the sheet and chunk detection step and calculates one label image corresponding to the detected explosive. It will be appreciated that the method described in connection with FIG. 5 can include more than two distinct detection paths according to the number of types of objects to be identified.

  Throughout this specification, the term “three-dimensional image” and the symbol C (i, j, k) are used to represent a set of CT slice images. The size of each CT slice is I columns × J rows. The symbol I in C (i, j, k) represents a column index and ranges from 0 to I-1. Similarly, the symbol j represents the row index and ranges from 0 to J-1. There are K of these slices in the set. The symbol k represents one of these slices and ranges from 0 to K-1. The function C (i, j, k) is used to refer to or represent a particular CT density in this set and is the CT density value in the i th column and j th row of the k th slice. Means that. CT density is expressed as a positive integer having 0 (Hounsfield units) corresponding to the density of air and 1000 (Hounsfield units) corresponding to the density of water, but other integer values may be used as desired. Can do.

  The function C (i, j, k) can be regarded as a three-dimensional image having a width of I pixels, a height of J pixels, and a depth of K pixels. Each element of the three-dimensional image is a voxel. The value C (i, j, k) of a specific voxel indicated by the (i, j, k) triplet is the CT density of the material occupying that voxel.

  The size of the voxel is determined by the resolution of the CT instrument. In one embodiment, the scanner has a nominal voxel size with a width (x) of 3.5 mm, a height (y) of 3.5 mm, and a depth (z) of 3.33 mm, which is a relatively small voxel. Thus producing a higher resolution compared to Eberhard et al. Systems, but the nominal size can be varied depending on several design factors. Using this information and CT density, it is possible to calculate the mass of each voxel in the three-dimensional image.

The CT density roughly corresponds to the physical density of the material. A CT density of 1000 corresponds to the density of water (ie 1 g / cc) to determine the mass of a given voxel in grams, so the CT density value for that voxel is divided by 1000 and voxel Multiplied by the volume (0.35 × 0.35 × 0.333 cc). The method described herein uses this conversion as (constant c _o ) to calculate the mass of the bag and the mass of each object identified in the bag.

  The main steps of the method of the present invention listed above and shown in FIG. 5 will now be described in detail. FIG. 6 shows a logic flow flow diagram of one embodiment of the problem area calculation 301 of the present invention. The goal of the problem calculation area is to remove the portion of the image that is outside the bag, thus reducing the data that other parts of the process analyze, speeding up the process and reducing memory requirements. In one embodiment, a rectangular subset containing all voxels with CT density values in the range of interest is extracted from the original image.

The problem area calculation input includes C (i, j, k), which is a three-dimensional CT image of the bag. The output is C _roi (i, j, k) representing the CT image of the bag region in question and the coordinates of the region of the box in question (x _min , x _max , y _min , y _max , z _min , z _max ) Including. The parameter used for the calculation is t _o , which is the air threshold for the bag. Method 301 first receives data C (i, j, k) representing the three-dimensional image of the bag and the value of air threshold t _o . Next, in step 312, voxels identified as containing data representing air are identified, and in step 314, the coordinates of the region of interest are calculated to exclude most if not all of the voxels. . Steps 312 and 314 proceed as follows to define the area of interest.

x _min = minimum value of I such that at least one C (i, j, k) ≧ t _o at any j, k x _max = at least one C (i, j, k at any j, k ) I maximum value such that ≧ t _o y _min = minimum value of j such that at least one C (i, j, k) ≧ t _o at any I, k y _max = any I, at least one C in k (i, j, k) ≧ t maximum value of _o become such j z _min = any I, at least one C at j (i, j, k) so as to be ≧ t _o The minimum value of k z _max = the maximum value of k such that at least one C (i, j, k) ≧ t _o at any I, j Next, the image of the problem area including the bag is represented by C _roi (I, j, k) = C (I + x _min , j + Y _min , k + Z _min ) where 0 ≦ I ≦ x _max −x _min
0 ≦ j ≦ y _max −y _min
0 ≦ k ≦ z _max −z _min

  FIG. 7 is a flow diagram illustrating the logic flow of one embodiment of the sheet detection method according to the present invention. Sheet explosives are characterized by one dimension (height, width or depth) that is much thinner than the other two dimensions. This dimension is called sheet explosive thickness. One sheet explosive detection method described herein can be tailored to sheet thickness and uses a constant false alarm rate (CFAR) method. CFAR's two-dimensional approach is described, for example, by Kreiten et al., "Discriminating Targets from Clutter" (Lincoln Lab Journal, Vol. 6, No. 1, 1993), Novak et al. “Target Recognition System” (Lincoln Lab Journal, Vol. 6, No. 1, 1993), and Frosgate et al. “Multiscale Segmentation and Anomaly Enhancement of SAR Imager”. ans.on Imag.Proc., Vol. 6, No. 1, 1997), all of which are incorporated herein by reference. In the three-dimensional CFAR approach of the present invention, as described in detail below, a CFAR sheet voxel analysis step 318 is performed on the CT image data of the problem area to identify which voxels are associated with the sheet-like object. A connected component labeling (CCL) method can then be applied to the sheet voxels at step 320, which can be connected within individual objects. In step 306, the objects are classified by mass distinction or the like.

  In the CFAR method of the present invention, each voxel in the bag is inspected to determine whether it is a part of a sheet-like explosive. To be part of a sheet explosive, a voxel must have a density value that is within a certain CT density value range and be statistically away from the background. In one embodiment, the background is a cube surface sized comparable to the sheet thickness centered around the test voxel, as shown in FIGS. 8A and 8B, which are schematic views of the preferred CFAR method of the present invention. Defined as a voxel in FIG. 8A shows in two dimensions a background cube 321 containing test voxels 319 applied to a CT data voxel containing a sheet-like object 317. Calculate the mean and standard deviation of the background voxels around the test voxel. Test voxel values are compared to background mean and standard deviation. If the statistical distance of the test voxel to the background is greater than a predetermined threshold, the test voxel is said to belong to a sheet explosive.

  In one alternative embodiment, not all of the voxels on the surface of the cube are used to calculate the mean and standard deviation. To save processing time, the surface voxels can be sampled and only the sampled voxels can be used in the mean and standard deviation calculations. In one embodiment, only every other voxel is sampled, resulting in the savings in processing time required to generate the mean and standard deviation.

  In another alternative embodiment, three separate two-dimensional CFAR calculations can be performed on three orthogonal Cartesian planes xy, xz, yz. The average and standard deviation of the background voxels is calculated for each plane, and the background is defined as the voxels around the squares of the individual planes. Next, the statistical distance of each plane is calculated and compared to a predetermined threshold. The threshold value may differ depending on the coordinate plane. The number of planes above the threshold is used to determine whether the voxel is a sheet voxel. For example, if one or more thresholds are exceeded, the voxel can be concluded as a voxel of the sheet. In another embodiment, a voxel labels a voxel on a sheet if more than one threshold is exceeded.

  In all cases of the CFAR method of the present invention, an upper threshold can be used in addition to or instead of the lower threshold. This removes the sheet-shaped object that shows a very large contrast with the background. An example of such a sheet is the outer surface of the bag.

  As shown in FIG. 8B, in the case of a massive object 325, the background voxels cover a larger portion of the object itself. Thus, the background is statistically closer to the test voxel selected as being within the test object. Thus, the CFAR distance is large for thin sheet objects and small for thick block objects. Using this property, all voxels belonging to the sheet-like object in the bag are detected and all voxels belonging to the massive object are deleted.

  According to the present invention, a three-dimensional CFAR method is applied to CT data to detect a sheet-like object. In developing the three-dimensional CFAR of the present invention, the selection of the size of the void area between the target and background voxels was considered. Like the processes described in the above-mentioned literature, standard prior art two-dimensional CFAR processes need to obtain background samples from areas that do not contain any part of the target. In the use of the present invention for detecting a sheet-like object, in the case of a sheet, only one dimension of the target object is known, and the direction of the dimension is unknown. Therefore, applying the prior art CFAR approach makes it difficult to sample the background, as is necessary with prior art CFAR processes. In order to realize the present invention, a part of the target object is also sampled as a background. In the present invention, the inclusion of a portion of the target sample in the background sample changes the mean and standard deviation of the background sample. However, this change is different for different target coverages of the CFAR sampling area. This is useful for the present invention to distinguish between sheet-like objects and massive objects. This difference between sheet coverage and lump coverage is shown in FIGS. 8A and 8B.

  After determining which voxels to consider as sheets, CCL analysis 320 is performed on the sheet-like voxels and the voxels are combined into a sheet-like object. The mass of each connected component thus obtained is compared with a predetermined mass threshold to determine the presence of a sheet-like explosive.

  Thus, the present invention uses the CFAR approach extended to three dimensions and significantly modifies and improves the two-dimensional techniques described in the literature. In the present invention, the modified CFAR is used as a step in the process to identify threats. The modified CFAR of the present invention first classifies individual voxels according to whether they are part of a sheet-like object. The process of the present invention then continues with an additional step, such as CCL, that combines voxels to target and a distinction step to determine whether the target poses a threat. In contrast, the previous two-dimensional CFAR approach was used as an independent detection algorithm, and its output consisted of the final classification of the object based on a two-dimensional CFAR analysis of the pixel of interest.

As described above, the purpose of the sheet-like explosive detection method is to detect a sheet-like object. A separate sheet explosive detection step is used to solve the problem of inadvertent removal of sheets from the data during morphological steps such as erosion performed during the mass detection process. The input of the sheet detection method includes C _roi (i, j, k), which is a three-dimensional image of the region in question (size I _roi × J _roi × K _roi ). The output of sheet explosive detection includes:
L _s (i, j, k) Sheet explosive label image (same size as _Croi )
N _{x ′} Number of detected sheet explosives ρ _{n ′} Density of each detection target M _{n ′} Mass of each detection target (x ⁿ _min , x ⁿ _max , y ⁿ _min , y ⁿ _max , z ⁿ _min , z ⁿ _max ) Boundary formation box for each detection target

The bounding box used in connection with this and other aspects of the invention is defined as the smallest rectangular area that contains the object to be bounded. The parameters for sheet detection are as follows.
(Ρ ^N _min , ρ ^N _max ) CT density range of the sheet problem g Size of the CFAR cube around the test pixel in the voxel t _l CFAR decision threshold Δ _s CCL threshold c _s CCL connectivity type (“plane ”,“ Edge ”or“ vertex ”)
Mass Threshold for Detection of M _s Sheet Explosives c _o CT Density to Mass Conversion Factor In one embodiment, the steps of the sheet explosive detection method include:

1. Start with a ROI 3D image C _roi (i, j, k).

2. On each side, the following embedded image is generated by embedding with a g-layer voxel value having some preset background value, such as zero.
P (i, j, k) Size (I _roi +2 g) × (J _roi +2 g) × (K _roi +2 g)

3. Raster scan the embedded image, with a CT density between ρ ^N _min and ρ ^N _max , voxels { _vo = ( _io , _jo , _ko ), _vl = ( _il , _jl , _kl ) ,. . . v _n . . . }. The shorthand notation v _n is used to indicate voxel 0 ≦ n <(I _roi +2 g) (J _roi +2 g) (Kroi +2 g).

Ｓ_n（立方体の表面積）のボクセル数は以下に等しい

4). For each voxel v _n = (i, j, k), the surface of the CFAR cube S _n is a (2g + 1) × (2g + 1) × (2g + 1) surface voxel v _nl = (I ′, j) centered on v _n ', K').

Number of voxels S _n (surface area of the cube) is equal to or less

5. The mean μ _n is calculated at the surface S _n of the CFAR cube centered on each v _n .

6). Calculate the standard deviation σ _n on the same surface.

7. Compute the distance d of the voxel v _n to the background given by μ _n and σ _n .

8). A CFAR image CFAR (i, j, k) having the same size as the input image _Croi (i, j, k) (zero is not embedded), and which is composed only of voxels whose distance d _n exceeds the threshold value t _1. Generate.

9. Using the CCL parameters, Δ _s and c _s , perform connected component labeling (CCL) on the CFAR image CFAR (i, j, k), label image L _s (i, j, k) and bounding box ( ^{_{x n min, x n max,}} y n min, y n max, z n min, z n max) to generate.

ここで、セレクタの関数ｈ（ｘ，ｌ）は下式のように定義される。

10. Each object l = 1,. . . For N _s , the mass M _l is calculated during CCL.

Here, the selector function h (x, l) is defined as follows.

11. Delete all objects whose mass M _l is below a given mass threshold m _s .

  12 Renumber the remaining objects using consecutive positive integer labels and update the label image. Set Ns equal to the number of remaining objects.

Regarding the sheet detection method, it should be noted that square deviation σ ² _n can be used in step 5 instead of standard deviation σ _n . As a result, the execution speed can be increased. Also, shapes other than cubes can be used to define the CFAR surface. Moreover, a thick sheet-like explosive can be detected by the lump of the method of the present invention. Thus, in one embodiment, the thickness of the detected sheet can be set slightly thicker than the thickest sheet that can be detected by mass detection.

  In another aspect of the invention, sheet-like objects can be detected with density data using a morphological approach similar to morphological CCL applied to mass object detection. In this morphological sheet detection approach, all objects in the data are eroded a predetermined number of times, so all thin sheet-shaped objects are deleted from the data. The number of erosion performed is based on the number of erosion required to remove the sheet-like object from the data, which is related to the thickness of the sheet. With each erosion, a layer of surface voxels can be removed. Thus, the number of erosion is related to the expected sheet thickness and voxel size. After performing all the erosion steps, the object remaining in the data is estimated as a massive object. The data associated with these objects is then deleted from further processing. Analyze the original data with the bulk object removed and label the sheet object. The remaining voxels are analyzed one at a time, such as with a CCL process, and the voxels are combined into a sheet-like object, and then a sheet-like object and a label are attached. Next, a distinction is performed on the sheet-like object, which is classified as a threat or non-threat, such as by comparing the object's mass with a predetermined mass threshold. A sheet with a mass above the threshold can be classified as a threat.

  Note that other sheet detection approaches, such as high pass filtering, can be used in various aspects of the invention to detect sheets. Also, for example, a connection process such as CCL can be applied to the binary data generated by sheet detection or the product of binary data and associated voxel density data.

  The bulk object detection process of the present invention can search for a bag image in a cluster of voxels in the density range of interest, label it as a bulk object, and use a mass-dependent density threshold to target Can be identified as a threat.

  In one embodiment, the mass detection process uses the CCL method to identify objects in a three-dimensional bag image. One of the main problems when using CCL is a composite object, ie two or more physical objects in close proximity that grow into one object in a bag image due to the finite resolution of the system. Creates a volume effect. To solve this problem, in one embodiment of the present invention, the image is preprocessed before applying the CCL to segment the composite object. This pre-treatment can be performed using an erosion operator, as will be described in detail below, which effectively removes the voxel surface layer from the object and allows the CCL to grow multiple objects together. To prevent. In order to balance the effect of erosion on the size of the object, the eroded image is segmented into multiple objects by CCL, and then an expansion operation as described in detail below is applied. This operation adds the surface voxels back to the object after it is determined that the object is a separate object.

  According to one aspect of the invention, the mass detection method uses one or more separate density ranges, one for each type of mass material in question having density values that fall within the range. In one embodiment, there are two distinct density ranges used for mass detection. The density range is selected according to the object to be identified. In one particular embodiment, one of the ranges deals with the first particular type of explosive (referred to herein as “Type A”) and the other one is all solid mass explosions except Type A. Including things. Because the two density range explosive types and typical false alarms are different, separate detection paths with different erosion and expansion can be used to optimize the performance of each density range.

In one embodiment, a given density range detection process comprises the following steps.
1. Perform at least one erosion to remove surface voxels.
2. CCL is used to segment the image into separate objects.
3. An inflation operator is used to recover the surface voxels.
4). Calculate the characteristics of the object.
5. A mass threshold is used to distinguish explosives from non-threat objects.

  Once the bag image is segmented into multiple objects, characteristics such as mass, average density, erosion mass, and erosion average density are calculated for each object. A density-dependent mass threshold is used to distinguish between threat and non-threat objects. The number of potential explosive objects, their characteristics and their coordinates in the bag image are returned by the mass detection process.

The input to the lump detection process includes C (i, j, k), a 3D CT image of the bag. The output includes:
N _b Number of detected block objects L _b (i, j, k) Label images (x ^min _n , y ^min _n , z ^min _n ), (x ^max _n ) having the same size as C (i, j, k) , Y ^max _n , z ^max _n ) Coordinates of the bounding box of each object V _n Number of voxels in each object M _n Mass of each object V ^u _n Number of erosion of voxel of each object M ^u _n Erosion mass of each object Parameters used in the process include:
(Ρ ⁿ _min , ρ ⁿ _max ) Type A explosive density range (ρ ^pe _min , ρ ^pe _max ) Type A explosive erosion density range (ρ ^b _min , ρ ^b _max ) Mass explosive density range ( ρ ^be _min , ρ ^be _max ) Erosion density of massive explosives (ρ ^bd _min , ρ ^bd _max ) Expansion density range of massive explosives (ρ ^bm _min , ρ ^bm _max ) Density range of massive explosives combined Δ _m massive adjacent type a eroded density range needed to maintain the erosion number n _p type a voxel erosion number e _b mass density range of maximum erosion density difference e _p type a density range between target for explosives merger maximum density difference c _b CCL connectivity type [rho _l low density and high connecting voxels minimum number delta _b CCL adjacent voxels mass erosion density range needed to maintain the minimum number n _b lump voxels of the voxel cylindrical pair comprising a density threshold m _l type a explosives between density bulk explosives The volume of the mass threshold m _l low-density mass threshold m _h minimum number of voxels c _v voxels required to maintain the subject after mass threshold V _min image segmentation step of a high density bulk explosives bulk explosives

  FIG. 9 is a schematic flow diagram illustrating the logic flow of one embodiment of the massive object detection method of the present invention. In pre-processing steps 330 and 332, image voxels of the bag are received. In the embodiment shown in FIG. 9, the pre-processing steps involving object erosion are performed separately and in parallel. In particular, type A material pre-processing is performed in step 330 and mass pre-processing is performed in step 332. It will be appreciated that these preprocessing steps need not be performed in separate parallel steps.

In step 330, the original CT image erosion operator, Type A eroded density range ^{_{(ρ pe min, ρ pe max}} ) for e _p times, or iteratively applied to each image, at step 332, the mass erosion density range Apply ( _b ^be _min , ρ ^be _max ) e _b times or both. The first iteration is the original CT image, ie, C ^o (i, j, k) = C (i, j, k). Further iterations are generated by applying an erosion operator to each voxel. At each voxel (i, j, k), the erosion operator performs the following steps:

1. Check if the voxel belongs to one of the erosion density ranges in question, set its CT value to zero, and if not, proceed to the next voxel.

2. Examine the 3 × 3 × 3 neighborhood of the current voxel (i, j, k). The number N of voxels (i ¹ , j ¹ , k ¹ ) having CT values C _n (i ¹ , j ¹ , k ¹ ) in the same density range is counted as the current voxel (type A or chunk).

3. If the count is below the threshold value n _p or ^{_{n b, C nl (i,}} j, k) equally set to zero, otherwise keep the current voxel. The threshold is different for chunks.

Other than the above formulas are type A substances.
Standard morphological erosion maintains a voxel only if the neighborhood of the voxel matches a particular pattern of mask. Normally, only voxels with all 27 neighbors in the problem range are maintained. Standard erosion is described, for example, in Serra, J. et al. In Image Analysis and Mathematical Morphology (Academic Press, London, 1982). Note that the erosion operator used in the morphological CCL of the present invention may be the voting or counting operator described above. Other erosion operators such as those described in Serra can also be used.

  In the process of the present invention, the purpose of the erosion operation is to separate multiple objects in close proximity to each other so that they can be merged together by standard CCL. In the present invention, the outer surface of the object is eroded. In contrast, some objects have internal holes (such as a thin cylindrical axial gap in a cylindrical bar object) or voxels that are outside the density range of interest due to image noise or artifacts. These erosion of the inner surface can divide the object into several parts or, in the case of thin objects with axial voids, can completely eliminate the object.

  The outer surface of the object is usually convex and the surface of the internal void is usually concave. Accordingly, the outer surface voxels have fewer target voxels in the vicinity of 3 × 3 × 3 than the inner surface voxels. Using the counting threshold, only the outer surface of the object can be selectively eroded while leaving the interior of the object.

  The results of the two preprocessing steps 330 and 332 are combined in an optional combination step 334. In the illustrated embodiment using a separate path for preprocessing, each preprocessing step 330 and 332 generates a unique set of preprocessing data from the original bag image data. In a combination step 334, these individual data sets are combined into one set of preprocessed data.

In segmentation step 335, the CCL is used to identify and label the eroded CT image data object. Neighboring voxels in the same density range are connected and assigned a target label for both the chunk (ρ ^b _min , ρ ^b _max ) and type A (ρ ^b _min , ρ ^b _max ). A neighboring voxel is defined as a combination of “face”, “edge” or “vertex” determined by the CCL connectivity parameter c _b .

Once the CCL process is executed, objects in both density ranges are selected as long as the distance between the objects is greater than the maximum density difference for connecting voxels with CCL, that is, ρ ^b _min −ρ ^b _max > Δ _b. Can be found. Running a CCL only once and comparing voxels to multiple density ranges is much more efficient and less time consuming than running a separate CCL for each range. This is because CCL generally uses a large amount of processing resources. If the threshold difference Δ _b is chosen to be smaller than the gap between the density ranges, there is no possibility of labeling a voxel of a substance in one of the ranges wrongly as belonging to another range, and thus multiple ranges Can run processes simultaneously. This approach, it will be understood that can be extended to any number of ranges that are separated by a gap with a size of at least a threshold delta _b.

A label image is generated, where each voxel (i, j, k) is assigned to the following values:

0 Voxel does not fall in any density range 1 ≦ n ≦ N _b Voxel belongs to the nth object where N _b is the total number of objects found in the bag image. Re labeling path CCL, calculating erosion speed V ^e _n voxels, the mass M ^c _n was eroded for each subject n.

In pruning step 336, the target that is composed of only a small number of voxels (V ^e _n <V _min) it is discarded, sets corresponding voxel L _b in the label image (i, j, k) to zero. The total number of objects _Nb is decreased, the remaining objects are renumbered with successive labels, and the label image is updated with new object labels.

A Type A expansion step 338 and / or a mass expansion step 340 can now be performed. As with the pretreatment erosion steps 330 and 332, the dilation steps 338 and 340 can be performed in separate parallel paths. Inflation adds a layer of voxels to the remaining objects to recover the mass and volume lost to erosion. For each voxel with a label attached (l = L _b (i, j, k)> 0), the erosion density l of the label application target including the current voxel is given by the following equation.

Here, M ^e _l and V ^e _l are calculated during CCL. This density is used to determine if the voxel is part of a bulk explosive object or a type A object. CT density C (i ¹ , j ¹ ) of all non-labeled voxels (L _b (i ¹ , j ¹ , k ¹ ) = 0) in the vicinity of 3 × 3 × 3 of the current voxel (i, j, k) , K ^l ). When the density is within the expansion density range corresponding to the current explosive class (lumps or type A), the following equation is obtained.

Next, the same label as the voxel (i, j, k) is assigned to the voxel (i ^l , j ^l , k ^l ).

The total number of voxels V _l and the total mass M _l are incremented. Expansion, e _p times in the density range of types A, e _b times the density range of the mass, can be performed continuously.

  Next, a second combination step 342 can be performed with the expanded data. In the illustrated embodiment using separate paths for Type A and mass inflation, each inflation step 338 and 340 generates a unique set of inflation data from eroded, segmented and pruned bag image data. . In a combination step 342, these individual data sets are combined into a set of dilation data.

The partial volume correction step 334 can now be performed. The mass of each object is enhanced by replacing the measured CT density of the voxels added by expansion with the average density of the object.

の場合、対象を合併して、質量Ｍ＝Ｍ_n＋Ｍ_mおよびボクセル数Ｖ＝Ｖ_n＋Ｖ_mの新しい１つの対象にすることができる。合併した対象が同じラベルを有するよう、ラベル画像Ｌｂ（ｉ，ｊ，ｋ）を更新する。棒形対象の検出の分析にこの合併を使用することについて、以下で詳細に説明する。 The partial volume correction 344 is described in detail below.
The target merger step 346 can now be executed. If the bounding boxes of objects n and m overlap and their erosion density is close, that is

In this case, the objects can be merged into one new object with mass M = M _n + M _m and voxel number V = V _n + V _m . The label image Lb (i, j, k) is updated so that the merged objects have the same label. The use of this merger to analyze the detection of bar objects is described in detail below.

の場合、対象は脅威である
この密度依存の質量閾値について、以下で詳細に説明する。 Next, a discrimination 308 can be performed. For each target n with 1 ≦ n ≦ N _b , determine whether this target is a potential threat based on the mass of gymnastics and the erosion density.

In this case, the target is a threat. This density-dependent mass threshold is described in detail below.

The connected component labeling (CCL) approach used in the present invention is used to detect objects with density values in a predetermined range of interest in a three-dimensional image. CCL is the process of determining whether a voxel belongs to a specific object. The objects are connected topologically (combination of “face”, “edge” and “vertex” connectivity determined by parameter c), and the voxel CT density value is within the problem range of ρ _l to ρ _b Yes, defined as a set of voxels in which the difference in voxel values between neighboring voxels is within a delta value Δ.

  A three-dimensional image is represented by a C (i, j, k) array. The index 0 ≦ k <K is the number of slices (index along the Z axis), and the value of K is determined by the length of the bag. The index 0 ≦ j <J is the number of rows (index along the Y axis), and in one embodiment, J = 117 or 158 depending on the shape of the bag (rectangular or square). The index 0 ≦ I <I is the number of columns (index along the X axis), and in one embodiment, I = 214 or 158, depending on the shape of the bag (rectangular or square).

Using an array L (i, j, k) of the same size as C (i, j, k), label the voxels. Either the background (labeled with L (i, j, k) = 0) or the number of objects n> 0 (labeled with L (i, j, k) = n). A label equivalent array l (I) is used in the CCL algorithm and a unique label is assigned to each target. This array must be initialized with l (I) = I, 0 ≦ I ≦ L _max , where the constant L _max is determined by the maximum number of distinct objects that can be seen in the bag.

Neighboring voxels (i _n , j _n , k _n ) (| I− _n | ≦ 1, | j−j _n | ≦ 1, | k−k _n | connected to the current voxel (i, j, k) Three different topological definitions of ≦ 1) can be used for the three-dimensional CCL process.
1. Voxels (i, j, k) and (i _n , j _n , k _n ) share a common plane | I−i _n | + | j−j _n | + | k−k _n | = 1.
2. Voxel (i, j, k) and _{(i n, j n, k} n) , the common edge _{| I-i n | + |} j-j n | + | share = 2 | k-k _n.
3. Voxel (i, j, k) and _{(i n, j n, k} n) , the common vertex _{| I-i n | + |} j-j n | + | share = 3 | k-k _n.

  These connectivity type combinations can be designated as input parameters for the CCL process. Note that if the current voxel (i, j, k) is on the surface of the bag image (I, j or k is equal to 0 or its respective maximum), not all possible neighboring voxels exist. I want.

  FIGS. 10A and 10B contain pseudo code describing one embodiment of the CCL method of the present invention. The first step of the CCL method is raster scanning of image data. This assigns a spare label to the voxel. If the current voxel is within the density range of interest, test against its neighboring voxels. If the value of the neighboring voxel is within the Δ range of the value of the current voxel, these two voxels are part of the same object. Since the first pass is executed sequentially, it is possible to assign different label numbers to different parts of the same object. When a merge point is encountered, the label equivalent is entered into the l (I) array. Use the lowest possible equivalent label and avoid circular references.

The second step of the method is to resolve all labels equivalent l (I) array, counts the number Nb of distinct target assigns each label to an equivalent value in the range of 1 to N _b. In the third step, for all (i, j, k), voxel labels are replaced with the equivalent label value L (i, j, k) = 1 (L (i, j, k)). During this pass, necessary information about each object can be accumulated, such as the bounding box index and the total mass of the object.

  Partial volume correction and mass enhancement of the object addresses the problem of density degradation of the surface voxels of the scanned object. The CT value for a given voxel represents the average density of objects within that voxel. As a result, accurate measurement of object density can only be achieved if the voxel is completely contained in the object. On the other hand, when a voxel is partially occupied by an object, its density degrades based on the portion occupied by the voxel. FIG. 11 shows the partial volume effect on the scanned object 351. In FIG. 11, voxels (squares) without diagonal lines are affected by the partial volume effect. The partial volume correction of the present invention is based on replacing the CT value of the scanned surface voxel using its average CT density or average of its erosion CT density.

Due to the partial volume effect, the actual density and mass of the scanned object is greater than its measured value. The difference in values depends on the shape of the object, in particular on the surface-to-volume ratio of the scanned object. A control with a large surface to volume ratio, such as a sheet or a small diameter cylinder, has a greater effect than a mass object or a small surface to volume ratio object, such as a large diameter cylinder. The measured CT density of the scan object is defined as the average density of all the voxels.

ここでＶはボクセルの体積、Ｃは質量変換定数である。 Here, ρ _v is the v-th CT density of the object, O is the object to be scanned, and V is the number of voxels to be scanned as viewed in CCL. Thus, the mass of the object is as follows.

Here, V is the volume of the voxel, and C is a mass conversion constant.

The target voxel can be divided into surface voxels and core voxels. A surface voxel is a voxel whose neighborhood of a scanning object voxel is less than 26. More accurate density measurements can be calculated using only the core voxels to be scanned. The new density is known as the erosion density.

Here C _o is the core of the scanned object, the N _e is erosion number of voxels scanned. Surface voxel number N _s is given by N-N _e. When the S _o and the set of surface voxels, the mass can be written as the following equation.

The degree of mass reduction due to the partial volume effect is determined by the second term on the right hand side of Equation 23. This is because ρ _v is most affected by the surface of the object. One way to correct it is to use an average or erosion density to replace ρ _v with the second term on the right hand side of Equation 23. The corrected mass can be calculated using one of the following equations.

  The erosion density or average density can be selected to replace the CT value of the surface voxel. However, the optimal substitution can be determined based on controlled experiments. Both densities can be used to compare the corrected mass to be scanned with its actual mass and select the density that produces the least error. An iterative algorithm can also be used to calculate the combination of erosion density and average density that produces a mass correction with the least dependence on the shape or size of the object.

  The merge process 346 described above in connection with FIG. 9 will now be described in detail. An erosion stage can be performed prior to the CCL process to avoid different object combinations within the scanned bag. As a result, different labels can be assigned to physically distinct objects. However, certain explosive devices, such as certain rod-shaped explosive objects, can be configured by bundling separate explosive objects together. With erosion, each object may be considered separately, thus losing the mass threshold criteria and reducing the overall detection rate. Each of these objects itself meets the density criteria, but its mass is below the threshold. In the present invention, a merge process can be performed based on the erosion density of the segmented object to recover the object.

  The data show that these objects have approximately the same erosion density as seen in the CCL process. The bounding box is in the same XYZ region. In one embodiment, this information leads to the conclusion that these objects must have the same one object component.

２．密度の一致が見られると、以下の近接測定の１つを計算する。
（Ｉ．）境界形成ボックスの近接性
（ｉｉ．）境界形成ボックスの中心間距離。距離は、ＸＹＺ空間、ＸＹ面、ＸＺ面またはＹＺ面で計算することができる。
３．対象が近接性判定基準を満たす、つまり距離が閾値より小さい場合は、２つの対象を合併して１つにする。
４．合併が不可能になるまで繰り返す。 In one embodiment, the merge approach only deals with objects that are within a specific density range that does not meet the mass threshold criteria. As a result, the following steps can be performed for each such object.
1. Search for objects of similar density. In other words, the following formula is obtained.

2. If a density match is found, calculate one of the following proximity measurements:
(I.) Proximity of bounding box (ii.) Center-to-center distance of bounding box. The distance can be calculated in XYZ space, XY plane, XZ plane or YZ plane.
3. If the object satisfies the proximity criterion, that is, if the distance is smaller than the threshold, the two objects are merged into one.
4). Repeat until merger becomes impossible.

  It should be noted that other criteria can be used to determine whether objects should be merged with each other, such as the position, shape and / or size of the objects.

  In general, the first step of the explosive detection process of the present invention is to detect objects within a given density range. This density range covers the vast majority of objects that are or may be seen in travel luggage. The second step is to delete most of these objects based on additional information such as mass. For example, an explosive may be considered a threat only if the mass is above a certain mass threshold.

The mass threshold can be used to reduce the false alarm rate. Increasing the mass threshold reduces false alarms and decreases detection rate. On the other hand, when the mass threshold is lowered, both the detection rate and the false alarm rate increase. As a compromise, in one embodiment, different density objects are subject to different threshold masses. The mass threshold can now be described below.

  According to the present invention, it is recognized that different density ranges are associated with different false alarm rates. As a result, selection of mass thresholds based on density ranges can be used to adjust the false alarm rate within a specific density range, thereby adjusting the overall system false alarm rate. For example, certain density ranges are associated with relatively high false alarm rates. A relatively high mass threshold can be selected for the density range to reduce the false alarm rate for that range. As a result, the overall system false alarm rate is reduced.

Other density dependent threshold masses can be used. For example, the density range can be divided into three or more non-overlapping regions. Furthermore, the step function can be changed so that the mass gradually changes between the two density regions.

Where M _T ¹ and M _T ² are the threshold masses of each region, ρ _T is the boundary between the two regions, and λ is the transition coefficient, which determines the width of the transition region between the two threshold masses. The transition factor λ can be automatically determined using a set of scanned travel items. FIG. 12 is a schematic plot of mass threshold versus density showing three different density dependent mass thresholds.

２．Ｎ_sおよびＮ_b双方がゼロでない場合、ラベル画像を融合する。

In one embodiment, the decisions made by the two detection methods, sheet and chunk, are arbitrated to arrive at a decision that all match. Therefore, one label file is obtained by fusing the two label images generated in the two parts of the process. Inputs to the decision data fusion method include:
N _s Total number of sheet explosives detected L _s (i, j, k) Label image of sheet explosives N _b Total number of detected massive explosives L _b (i, j, k) The label image output includes:
N Total number of explosives detected L (i, j, k) Fused label image In one embodiment, the decision data fusion process includes the following steps.
1. Check the number N _{s of} sheet explosives detected and the number N _b of massive explosives detected. Both must not be zero in order to perform data fusion. Otherwise, obtain a non-zero explosive label image.

2. If both N _s and N _b are not zero, the label images are fused.

  If a voxel has two conflicting labels, one of which designates it as part of a sheet object and the other labels it as a block object, the voxel block object label is used, so any Note that this decision is made. Also, if the output specification of the decision method changes, the data fusion portion of the method can be turned off to make the sheet-like object and the massive object into two separate output label images.

  Several variations and additions can be incorporated into the determination process of the present invention. Next, this will be described in detail.

  Processing time can be shortened by first detecting a lump and subtracting the detection target from the image. The lump detection takes less time than the sheet detection. Next, the detection of sheet-like explosives works only with the remaining voxels, speeding up the overall detection process. That is, in one embodiment, detection is performed in multiple stages so that the overall detection process is more efficient. Each item that can be identified by the method of the present invention is generally associated with a unique set of detection steps. In the present invention, a detection approach is applied to eliminate inefficiencies introduced by iteratively processing data with multiple detection methods. When one specific detection procedure is applied to a set of data and used to classify that data, the data is removed from further processing.

  Terminating the detection process after detecting the first explosive also increases system throughput. This is because a potentially explosive object is found or the operator inspects the bag image.

  Execution time limits can also be applied to the process. The run time of the detection process is greatly increased for bags with a large total volume occupied by dense objects. Such travel items can be declared suspicious and forwarded for further inspection by the operator without having to complete all the steps of the detection process.

  The sheet detection approach can be simplified using measures to merge the divided sheets by extending the surface plane. Sheet explosives may be detected as smaller pieces when located near relatively opaque objects such as metal bars. If not merged, each piece is deleted by the mass threshold. One way to merge these pieces according to the present invention is to fit a face to each piece and determine the intersection of the two faces. If the intersection is close to both pieces (within predetermined limits), both pieces are considered as parts of a larger sheet and the quality Y braze is the total mass of the pieces. As a result, the merged object exceeds the mass threshold and is classified as a threat object.

  It may be desirable to detect a liquid object. Once it is determined that the liquid object should be classified as a non-threat item, it is advantageous to identify the liquid as such to reduce the number of alarms that can occur.

  Many bags contain liquid bottles (shampoo, water, wine, etc.). If these liquid bottles cause false alarms, this can be distinguished using the liquid detection method of the present invention, thus reducing the number of false alarms. Liquid detection methods distinguish liquid bottles from fixed objects. This detection takes advantage of the fact that the surface of the liquid is at the same level as horizontal, and that there is usually air above the surface. This assumes that the liquid bottle is not completely filled. Given the bounding box and voxel to be detected, determine the number of voxels that touch each surface of the bounding box and calculate the percentage of the top count in the total count. It is concluded that a voxel is labeled with a portion of the object, i.e. a liquid bottle, and touches the surface of the bounding box when it is at or near the surface of the bounding box. If the percentage is greater than a predetermined threshold and the average value of voxels above the liquid surface is close to the value of air, the object is liquid.

  In one embodiment, the present invention first determines whether the object is contained liquid by creating a bounding box surrounding the object. A histogram of the number of top voxels and a histogram of the number of bottom voxels are calculated along the height of the bounding box. Here, the top voxel is defined as the first voxel to visit while traversing the row of voxels in the bounding box from top to bottom. The bottom voxel is defined as the last visited voxel while traversing the row of voxels in the bounding box from top to bottom. The position of the maximum value of the histogram of the top voxels defines the position of the top surface along the height of the bounding box, and the maximum count of the histogram defines the number of top surface voxels. Similarly, for a bottom histogram, the position of the maximum value defines the position of the bottom surface, and the maximum count defines the number of bottom surface voxels. Calculate the ratio of the number of top voxels to the top surface area of the bounding box. This ratio exceeds a predetermined threshold, the ratio of the number of upper surface voxels to the number of lower surface voxels exceeds another predetermined threshold, and the average density of voxels above the upper surface indicates that air is located above the upper surface. If so, it is concluded that the subject is a contained liquid. In one embodiment, this can be concluded that the subject does not pose a threat.

The overall performance of the system can be optimized, including detection rate and false alarm rate. The overall detection rate for many types of explosives depends on a priori possibilities and the likelihood of detection. The likelihood of detection depends on the individual detection process for each such explosive type. The detection process can be performed very successfully with one type, but not very well with another type of explosive. In one embodiment, the overall detection rate is an average of the individual detection rates. The overall false alarm rate of the system also depends on the individual false alarm rate. In one embodiment, the overall system false alarm rate is the sum of the individual false alarm rates. For example, the individual and overall system detection rates and false alarm rates may be determined for a particular set of three different types of substances labeled Type 1, Type 2 and Type 3 for illustrative purposes. The thresholds and parameters can be expressed in Table 1.

  According to the present invention, it is recognized that the individual detection rates of the three substance types are correlated. That is, for example, the detection rate of a type 2 substance is affected by the detection rate of a type 1 substance. As the detection rate changes, the corresponding false alarm rate also changes.

For example, if a set of system specifications requiring an overall system detection rate of P _D ≧ 95% and an overall system false alarm rate of P _FA ≦ 10% is applied to the system in Table 1, Does not meet requirements. According to the present invention, one or more of the individual detection rates can be adjusted to adapt the overall system. This can be done, for example, by reducing the detection rate of type 2 and type 3 substances. An example of the result of this adjustment is shown in FIG. It can be seen that the values in Tables 1 and 2 are used for illustrative purposes only and do not reflect actual system parameters.

  As shown, as a result of adjusting the two individual detection rates, the system now meets the sample specifications. It can be seen that as each detection rate is adjusted down, the associated false alarm rate also decreases. Thus, given a desired range of overall and individual detection rates, the overall detection rate and / or false alarm rate can be optimized by adjusting the likelihood of individual detection and the false alarm rate. it can. In one embodiment, the system can be tuned for a specific detection rate and / or a specific false alarm rate.

  Note that individual detection rates may depend on other individual detection rates. As a result, adjusting one rate may inadvertently change another rate. To clarify this effect, statistical data analysis approaches such as simulated annealing and general algorithms are used to adjust individual rates as needed for specific desired overall system performance. The necessary parameters can be determined.

  In the present invention, the detection rate can be adjusted one or more in several approaches. For example, comprehensive analysis of actual threats and non-threat items resulted in a relationship between target density and threat mass threshold. This allows the threshold to be adjusted for a particular density and thus for a particular threat item. If it is desirable to reduce the detection rate of a particular threat and thus reduce the false alarm rate, the mass threshold can be increased for a specific threat density range. In addition to or instead of adjusting the mass threshold, other such deterministic parameter adjustment approaches can be used. This means that the other parameters detailed in this specification can be adjusted to adjust the individual detection rate and / or false alarm rate, thereby adjusting the overall system detection rate and / or false alarm rate. can do. In addition to or instead of these deterministic approaches, using statistical approaches such as simulated annealing and general algorithms applied to data obtained on real threats and non-threat objects, One or more parameters and / or thresholds can be adjusted to adjust the detection rate and / or false alarm rate.

  Intelligence data can provide additional information regarding the possible shapes of the explosive device and the likely location within the bag. Statistics on the shape and location of typical non-threat items carried in checked travel items can be collected in item testing. This information can be used in making a method decision to further distinguish between threats and harmless targets. For example, new distinguishing features and / or process changes may be incorporated to reveal perishable products.

  While the invention has been particularly shown and described with respect to preferred embodiments thereof, those skilled in the art will recognize that various changes and modifications can be made without departing from the spirit and scope of the invention as defined in the claims. Is understood. For example, the present invention can be applied to detect explosives other than the specific substances disclosed above. The present invention can also be applied to the detection of other objects and substances such as drugs and currency. The present invention can be used to detect checked and carried travel items and these items in transport and other types of containers.

1 is a perspective view of a baggage scanning system according to the present invention. FIG. FIG. 2 is an end cross-sectional view of the system shown in FIG. FIG. 2 is a radial cross-sectional view of the system shown in FIG. 1. 1 is a schematic electrical and mechanical block diagram of one embodiment of a baggage scanner of the present invention. FIG. FIG. 4 is a top level flow diagram illustrating the logic flow of one embodiment of the object identification method of the present invention. FIG. FIG. 4 is a flow diagram of the logic flow of one embodiment of the subject calculation domain of the present invention. FIG. 4 is a flow diagram of the logic flow of one embodiment of a sheet detection method according to the present invention. It is the schematic of the sheet-like object detection method of FIG. It is the schematic of the sheet-like object detection method of FIG. 4 is a flow diagram of the logic flow of one embodiment of a massive object detection method according to the present invention. FIG. 3 is pseudo code describing one embodiment of a modified connection component labeling method according to the present invention. FIG. FIG. 3 is pseudo code describing one embodiment of a modified connection component labeling method according to the present invention. FIG. 1 is a schematic illustration of a partial volume effect. FIG. 3 is a schematic plot of mass threshold versus density representing three different density dependent mass thresholds in accordance with the present invention.

Claims (24)

A method for detecting a liquid represented by computed tomography (CT) data of an area, comprising:
Identifying a plurality of volume elements with CT data associated with the subject, each volume element associated with a density value;
Defining a sub-region surrounding the object, wherein the sub-region defines a plurality of sub-region surfaces;
Identifying the surface volume element of interest on the surface of the subregion;
Identifying the upper surface of the lower region;
A method comprising labeling a subject as containing liquid if the number of surface volume elements proximate to the upper surface of the subregion exceeds a predetermined threshold portion of the total number of surface volume elements. An apparatus for detecting liquid with computed tomography (CT) data of an area,
Means for identifying a plurality of volume elements in the CT data associated with the object, each volume element associated with a density value;
Means for defining a subregion surrounding the object, the subregion defining a plurality of subregion surfaces;
Means for identifying a surface volume element of interest on the surface of the subregion;
Means for identifying the upper surface of the lower region;
Means for labeling the object as containing liquid if the number of surface volume elements proximate to the upper surface of the lower region exceeds a predetermined threshold portion of the total number of surface volume elements. A computed tomography (CT) scanning system for identifying an object in an area, comprising:
Means for collecting CT data of the region;
Means for identifying a plurality of volume elements with CT data associated with the object, each volume element associated with a density value;
Means for defining a subregion surrounding the object, the subregion defining a plurality of subregion surfaces;
Means for identifying a surface volume element of interest on the surface of the subregion;
Means for identifying the upper surface of the lower region;
Means for labeling the subject as containing liquid if the number of surface volume elements proximate to the upper surface of the subregion exceeds a predetermined threshold portion of the total number of surface volume elements. A method for identifying an object represented by a computed tomography (CT) data of an area, comprising:
Identifying a plurality of volume elements of the region in the CT data, each volume element associated with a density value;
Identifying a first density range and a second density range;
For each of the plurality of volume elements, comparing the density value of the volume element to at least one of the density ranges;
If the density value of the volume element falls within one of the density ranges, the method includes labeling the volume element as related to the subject in question. An apparatus for identifying an object represented in a computed tomography (CT) data of an area,
Means for identifying a plurality of volume elements of the region in the CT data, wherein each volume element is associated with a density value;
Means for identifying the first density range and the second density range;
Means for comparing the density range of the volume element with at least one of the density ranges;
Means for labeling the volume element as related to the object in question if the density value of the volume element falls within one of the density ranges. A computed tomography (CT) scanning system for processing CT data of a region,
Means for collecting CT data of the region;
Means for identifying a plurality of volume elements of the region in the CT data, each volume element being associated with a density range,
Means for identifying the first density range and the second density range;
Means for comparing the density value of the volume element with at least one of the density ranges;
And means for labeling the volume element as related to the object of interest if the density value of the volume element falls within one of the density ranges. A method for processing computed tomography (CT) data of an object in an area, comprising:
Identifying a plurality of volume elements associated with the object, wherein each volume element is associated with a density value;
For each volume element in question, (i) identify multiple neighboring volume elements in the region; (ii) compare the density value associated with each neighboring volume element with a predetermined target density range; and (iii) Count the number of neighboring volume elements within a predetermined target density range; (iv) compare the number of neighboring volume elements whose density range is within a predetermined target density range with a predetermined threshold; and (v) density range Removing the volume element from the object if the number of neighboring volume elements that are within the predetermined object density range does not exceed a predetermined threshold. An apparatus for processing computed tomography (CT) data of an object in an area, comprising:
Means for identifying a plurality of volume elements in the CT data, wherein each volume element is associated with a density value;
Means for identifying a plurality of neighboring volume elements of the region for each volume element in question;
Means for comparing the density value associated with each neighboring volume element with a density range of a given object;
Means for counting the number of neighboring volume elements whose density values fall within a predetermined target density range;
Means for comparing the number of neighboring volume elements whose density values fall within a predetermined target density range with a predetermined threshold;
Means for removing volume elements from the object if the number of neighboring volume elements whose density values fall within the density range of the predetermined object does not exceed a predetermined threshold. apparatus. A computed tomography (CT) scanning system for processing CT data of a region,
Means for collecting CT data of subjects in the area;
Means for identifying a plurality of volume elements associated with the object in the CT data, wherein each image volume element is associated with a density value;
Means for identifying a plurality of neighboring volume elements of the region for each volume element in question;
Means for comparing the density value associated with each neighboring volume element with a density range of a given object;
Means for counting the number of neighboring volume elements whose density values fall within a predetermined target density range;
Means for comparing the number of neighboring volume elements whose density values fall within a predetermined target density range with a predetermined threshold;
Means for removing volume elements from an object if the number of neighboring volume elements whose density values fall within the density range of the predetermined object does not exceed a predetermined threshold. A method for detecting a sheet-like object represented by computed tomography (CT) data of a region, comprising:
Identifying a plurality of volume elements in the CT data, wherein each volume element is associated with a density value;
Performing an erosion step for removing volume elements from the CT data of the region a predetermined number of times N, wherein the erosion step generates erosion CT data;
Identifying the eroded object in the erosion CT data;
Removing eroded volume elements from the CT data;
A method comprising labeling an object as a sheet-like object with CT data after removing the erosion object. An apparatus for detecting a sheet-like object represented in a computed tomography (CT) data of an area,
Means for identifying a plurality of volume elements in the CT data, wherein each volume element is associated with a density value;
Means for performing a predetermined number N of erosion steps to remove volume elements from the CT data of the region, said erosion step generating eroded CT data;
Means for identifying the erosion object in the erosion CT data;
Means for removing volume elements to be eroded from CT data;
An apparatus comprising means for attaching a label to a target of CT data as a sheet-like target after removing the target of erosion. A computed tomography (CT) scanning system for processing CT data of a region,
Means for collecting CT data of the region;
Means for identifying a plurality of volume elements in the CT data, wherein each volume element is associated with a density value;
Means for performing an erosion step for removing volume elements from the CT data of the region a predetermined number of times N, wherein the erosion step generates erosion CT data;
Means for identifying the erosion object in the erosion CT data;
Means for removing volume elements to be eroded from CT data;
A system comprising means for labeling a target of CT data as a sheet-like target after removing the target of erosion. A method of processing computed tomography (CT) data representing an object within a region comprising:
Identifying a plurality of volume elements with CT data associated with the subject, each volume element associated with a density value;
Identifying a surface volume element of the object, wherein the surface volume element of the object is disposed on the surface of the object;
Calculating an average density of the object, wherein the average density is an average of density values of volume elements associated with the object;
Adjusting the density value of the surface volume element to an adjusted density value, wherein the adjusted density value is based on a function of average density. An apparatus for processing computed tomography (CT) data of an object within an area,
Means for identifying a plurality of volume elements in the CT data associated with the object, each volume element having a density value;
Means for identifying a surface volume element of the object, wherein the surface volume element of the object is disposed on the surface of the object;
Means for calculating an average density of the object, wherein the average density is an average of density values of volume elements associated with the object;
Means for adjusting the density value of the surface volume element to an adjusted density value, said adjusted density value being a function of the average density. A computed tomography (CT) system for processing CT data representing a region of an object,
Means for collecting the CT data of the subject;
Means for identifying a plurality of volume elements with CT data associated with the object, each volume element having a density value,
Means for identifying a surface volume element of the object, wherein the surface volume element of the object is disposed on the surface of the object;
Means for calculating an average density of the object, wherein the average density is an average of density values of volume elements associated with the object;
A system comprising means for adjusting a density value of a surface volume element to an adjusted density value, wherein the adjusted density value is a function of an average density. A method for classifying an object represented by a computed tomography (CT) data of an area, comprising:
Calculating the mass and density of the object;
Comparing the density of the object with a predetermined density range and the mass of the object with the predetermined mass range, wherein the predetermined mass range is related to the predetermined density range;
A method comprising classifying an object as belonging to a predetermined object class if the object's density falls within a predetermined density range and the object's mass falls within a predetermined mass range. An apparatus for classifying objects by computer tomography (CT) data in a certain area,
Means for calculating the mass of the object;
Means for calculating the density of the object;
Means for generating a predetermined density range;
Means for generating a predetermined mass range, wherein the predetermined mass range is associated with a predetermined density range;
Means for comparing the density of the object with a predetermined density range;
Means for comparing the mass of the object with a predetermined mass range;
An apparatus comprising: means for classifying an object as belonging to a predetermined object class when the density of the object falls within a predetermined density range and the mass of the object falls within a predetermined mass range. A computed tomography (CT) scanning system for classifying objects in a region,
Means for collecting CT data of the region;
Means for calculating the mass of the object using CT data;
Means for calculating the density of the object using CT data;
Means for generating a predetermined density range;
Means for generating a predetermined mass range, wherein the predetermined mass range is associated with a predetermined density range;
Means for comparing the density of the object with a predetermined density range;
Means for comparing the mass of the object with a predetermined mass range;
A system comprising means for classifying an object as belonging to a predetermined object class when the density of the object falls within a predetermined density range and the mass of the object falls within a predetermined mass range. A method for detecting an object represented by a computed tomography (CT) data of an area, comprising:
Identifying a plurality of volume elements in the CT data of the region, wherein each volume element is associated with a density value;
Performing a sheet-like object detection process on the CT data of the region to identify volume elements associated with any sheet-like object;
A method comprising connecting and targeting volume elements of CT data of a region after performing a sheet-like object detection process. An apparatus for detecting an object represented by computed tomography (CT) data in a certain area,
Means for identifying a plurality of volume elements in the CT data of the region, wherein each volume element is associated with a density value;
Means for performing a sheet-like object detection process on the CT data of the region to identify a volume element associated with any sheet-like object represented by the CT data of the region;
An apparatus comprising: a means for connecting a volume element of CT data in a region to be an object after performing a sheet-like object detection process. A computed tomography (CT) scanning system for detecting an object in an area, comprising:
Means for collecting CT data of the region;
Means for identifying a plurality of volume elements in the CT data of the region, wherein each volume element is associated with a density value;
Means for performing a sheet-like object detection process on region CT data to identify volume elements associated with any sheet-like object represented by region CT data;
And a means for connecting the volume elements of the CT data of the region to be the target after performing the sheet-like target detection process. A method for detecting an object represented by a computed tomography (CT) of an area, comprising:
Identifying a plurality of volume elements in the CT data, wherein each volume element is associated with a density value;
Analyzing the CT data to identify surface volume elements at the surface of the object represented by the CT data;
Removing surface volume elements from CT data such that multiple objects represented by CT data in close proximity to each other are separated from each other;
Combining volumetric elements of CT data into multiple separate objects;
Adding a replacement surface volume element to the object to restore the object to its original size. An apparatus for detecting an object represented by a computed tomography (CT) data of an area,
Means for identifying a plurality of volume elements in the CT data, wherein each volume element is associated with a density value;
Means for analyzing the CT data to identify surface volume elements at the surface of interest represented in the CT data;
Means for removing surface volume elements from the CT data such that multiple objects represented by the CT data in close proximity to each other are separated from each other;
Means for combining the volume elements of CT data into a plurality of separate objects;
Means for adding a replacement surface volume element to the object to restore the object to its original size. A computed tomography (CT) scanning system for identifying an object in an area, comprising:
Means for collecting CT data representing a region;
Means for identifying a plurality of volume elements in the CT data, each volume element associated with a density value;
Means for analyzing the CT data to identify surface volume elements at the surface of interest represented in the CT data;
Means for removing surface volume elements from the CT data such that multiple objects represented by the CT data in close proximity to each other are separated from each other;
Means for combining the volume elements of CT data into a plurality of separate objects;
Means for adding a replacement surface volume element to restore the object to its original size.

Images