CZ306898B6 - A method of detecting unauthorized access attempts to protected areas - Google Patents

Abstract

The method of detecting unauthorized access attempts to protected areas, wherein the space before entering the protected areas is sensed by a motion sensor adapted for measuring the depth map in the field of view of its camera illuminated by infrared light and the images are sent to the evaluation unit, comprising the following steps: a) firstly, a reference image of the space is created and sent without the presence of persons, and b) subsequently, for at least some of the other images in the evaluation unit, reading of the reference image is performed from them in order to obtain calculated images, c) if objects are detected in the images thus calculated, objects of an ellipsoidal shape are detected when the state is evaluated as an unauthorized access attempt, if either d) after a set of consecutive calculated images, comprising objects of an ellipsoidal shape, is followed by a set of at least one calculated image without objects and is followed by at least one calculated image with an object and/or e) in at least two consecutive calculated images, at least two objects of an ellipsoidal shape are detected.

Description

Technical field

The present invention relates to a method for detecting unauthorized passage attempts into protected areas, respectively. detecting the passage of an unauthorized person when passing through an authorized person.

BACKGROUND OF THE INVENTION

The basic function of the access control system is to control the access of persons to protected areas. The access control system may consist, for example, of an electronic part - card readers, biometric devices (eg hand shape detection) and necessary other electronics that control mechanical barriers - turnstiles, barriers, but especially doors. At the time the door is opened by another authorized person, another person or persons may enter through the door until the door is closed again, with or without the piggybacking of the authorized person who realized the door opening.

The following systems are known in the art to prevent or detect unauthorized passage of persons. Mechanical barriers (turnstiles, security portals), which, however, require active cooperation of the authorized person during the passage, are generally slow and expensive, and are also difficult to use in the case of disabled persons or when handling larger luggage. Another solution is to install the electronic sensor (s) on the door frame. Such a sensor emits an infrared beam and at the same time senses its reflection; if a person passes, the beam is interrupted. The disadvantage of this approach is its low resistance to unauthorized piggybacking as well as ensuring proper operation when handling larger luggage or trolleys. Alternatively, it is possible to evaluate the image data obtained from the camera surveillance system for the automatic detection of persons in front of the door unit. The disadvantage of this approach is high purchase costs and low durability in the case of more sophisticated forms of unauthorized piggybacking.

WO 2015/092 049 discloses a person recognition method in which the head and shoulders of a person are detected by analyzing the depth map in the field of view. This procedure is computationally demanding and is not effective in reduced visibility, respectively. in the dark. Other possible person recognition procedures are described in US 8 326 084, US 2012/087572, US 2014/139633 and WO 03073359, or in the Embedded based tailgating / piggybacking detection security system by Tjun Wem Chan et al., Published in Humanities, Science and Engineering (CHUSER), 2012. IEEC Colloquium on Kota Kinbalu, Malaysia.

SUMMARY OF THE INVENTION

The disadvantages of the prior art are eliminated by a method of detecting unauthorized passage into protected areas, in which the area before entering the protected area is scanned by a motion sensor adapted to measure the depth map in the field of view of its camera at infrared illumination, and images are sent to the evaluation unit. storing reference ranges of shapes and sizes of elliptical objects corresponding to the predicted ranges of shapes and sizes of human heads in images from the motion sensor, wherein

(a) a reference image of the absence of persons is first created and transmitted; and

b) subsequently, for at least some of the other images in the evaluation unit, the reference image is subtracted from them to obtain the calculated images,

(c) if objects are detected in such calculated images, objects of an elliptical shape falling by their shape and size within the reference ranges of elliptical shapes and sizes shall be detected

The object is evaluated as an attempted unauthorized passage if either

d) a set of consecutive calculated images containing said elliptical-shaped objects is followed by a set of at least 1 calculated image without objects, but no more than the number of calculated images without objects corresponding to a maximum of 10 seconds of space capture; object and / or

e) in at least two consecutive calculated images, at least two elliptical objects falling in shape and size within the reference ranges of elliptical objects and sizes stored in the evaluation unit are detected.

Preferably, for detecting objects in step c), d) or e), the images calculated in step b) are segmented using adaptive threshold intensity histograms.

To reduce false detection of unauthorized passage attempts, the condition in step e) is preferably evaluated as an unauthorized passage attempt if at least two elliptical objects falling in shape and size within the reference shape ranges and elliptical objects stored in the evaluation unit are detected in the evaluation unit. at least six out of ten consecutive calculated images.

It is also advantageous if, in step a), the maximum permissible position change of the object with elliptical contours in the immediately consecutive calculated images is stored in the evaluation unit, and in order to limit false detection of unauthorized passage attempts, the state in step e) is evaluated as an unauthorized attempt. pass if the position change of the elliptical shape in pairs of consecutive calculated images is less than or equal to said maximum allowed position change.

In step d), said set of consecutive computed images without objects preferably corresponds to images taken over a period of 1 to 10, more preferably 3 to 5 seconds.

In an alternative embodiment of the unauthorized passage detection method, the area before entering the protected area is scanned by a motion sensor adapted to measure the depth map in its camera field of view at infrared, and the images are sent to an evaluation unit in which the reference shape ranges are stored. and the size of the elliptical objects corresponding to the predicted ranges of shapes and sizes of the heads of the people in the images from the motion sensor, and the computing unit is coupled to a door opening sensor that sends an opening signal to the computing unit,

(a) a reference image of the absence of persons is first created and transmitted; and

b) subsequently, for at least some of the other images in the evaluation unit, the reference image is subtracted from them to obtain the calculated images,

(c) if objects are detected in such calculated images, elliptical objects falling in shape and size within the reference ranges of the elliptical objects and sizes stored in the evaluation unit are detected, the condition being evaluated as an attempted unauthorized passage if either

(d) a set of consecutive calculated images containing at least one said elliptical shape object followed by a set of at least 1 calculated image without objects and immediately followed by at least one calculated image with an object calculated from the image taken at the door opening period;

e) in at least two consecutive calculated images, at least two elliptical objects falling in shape and size within the reference ranges of elliptical objects and sizes stored in the evaluation unit are detected.

-2GB 306898 B6

The proposed procedure carries out the automatic detection of other persons who enter the demarcated area between the release of the mechanical barrier, such as the door, and its closure, outside of the person who performed the identification. If another person enters the monitored area in front of the door after opening the door, an alarm signal is triggered. Persons requesting to open the bar can carry normal hand luggage or handle the truck. In an exemplary embodiment, the algorithm operates with a depth map obtained using a Kinect sensor. Kinect is a sensor that measures the depth map in the field of view of a special camera with the help of infrared illumination (a special grid whose distortions allow to determine the distance of a certain point from the camera). Thus, the proposed method of analyzing a computed image of a guarded area processes depth information to detect unauthorized inputs.

The advantage of the proposed procedure is also independence from the lighting of the control station, where the sensor, which is based on infrared scanning, can be realized even in complete darkness. Furthermore, persons can carry hand luggage and the truck can also be handled. Furthermore, the proposed algorithm is resistant to various piggybacking attacks.

Due to the presence of IR component in solar radiation, the proposed algorithm can be used only in interiors.

Clarification of drawings

The invention is further explained in more detail by way of example embodiments and drawings, wherein Fig. 1 is a flowchart of an exemplary embodiment of the method of the invention; Fig. 2 is a flowchart of an exemplary head detection method (which is part of the method of the invention); Fig. 4 is the same image after areas outside the depth measurement range have been suppressed; Fig. 5a is the histogram of the motion sensor image; Fig. 5b is the same histogram after filtration Fig. 6a shows the histogram of Fig. 5b with the minima indicated, and Fig. 6b shows the histogram of Fig. 5b with the maxima indicated, Fig. 7 shows the image of Fig. 4 with the first segment corresponding to the object closest to the sensor. Fig. 8 is an example of a segmented image with distinct detections; Fig. 9 is a segmented image with the presence of one person in which the movement of the first segment of the Figure 10 shows a similar movement of two persons in the image.

DETAILED DESCRIPTION OF THE INVENTION

A motion sensor is mounted above the monitored area in front of the door to measure the depth map in the camera's field of vision at infrared illumination. The motion sensor is connected to the evaluation device.

It is advantageous to place a motion sensor for carrying out the method according to the invention above that part of the space in front of the door / guarded gate / mechanical barrier in which a person moves just before passing through said guarded door / guarded gate. The height above ground is chosen to cover most of the surrounding area where an unauthorized person could pass, for example, to occupy an area 5 times greater than the perimeter of the person's head in the space. Preferably, the height of the sensor above the floor is 200 to 270 cm.

Although designed as a Kinect motion sensor, it is clear that other motion sensors, such as Leap Motion or Google Tango, are also applicable.

After the motion sensor has been installed above the area of interest, a reference image (without the presence of persons) is taken and sent to the evaluation device and stored therein.

-3 CZ 306898 B6

Subsequently, all or at least some other images from the motion sensor are sent and evaluated.

An example of an image, ie a depth map from a motion sensor, is shown in Fig. 3.

A motion detector based on a background model is applied to the input image representing the depth map from the motion sensor. Thanks to the depth map, which is independent of the way of lighting, there is no need to use computationally complex algorithms in this case, but a simple difference between the current and the reference image is sufficient. If this difference is detected, possible objects representing one or more persons are selected (Fig. 4). baggage.

These objects are subjected to the following analysis based on the detection of elliptical head shapes in the depth map. First, a segmentation process is applied to the calculated image (representing the difference between the current image and the reference image), which uses depth map structures and proceeds from the areas closest to the motion sensor (camera) to the areas furthest from the motion sensor. The individual thresholds for the segmentation process are adaptively determined from the depth map histogram (Fig. 5a), which is first filtered to suppress abrupt changes and capture major luminance levels (Fig. 5b). Each significant level of the source computed image produces a minimum (Figure 6a) and a maximum (Figure 6b) in the histogram. The minimum and maximum positions in the filtered histogram generate brightness ranges that correspond to objects located at a similar distance to the motion sensor. Based on these ranges, segments are generated in which elliptical shapes are then detected. An exception is the first segment, which in the source data represents the object closest to the motion sensor (Fig. 7).

The grayscale motion sensor source data contains two key levels that are unimportant for processing and are therefore suppressed. This is the white and black level area, which means the area where the depth matrix (black) cannot be determined, and the area outside the depth measurement range (white). In a particular embodiment of the method according to the invention, to realize the segmentation, it is desirable to sprout a solid white color and replace it to solid black (change from Fig. 3 to Fig. 4).

The calculated image thus adjusted is segmented according to the histogram (Fig. 5a). The first step is to filter the histogram in the window to suppress sharp changes and capture major luminance levels. The result of simple filtering for a window of 20 elements (columns) is shown in Fig. 5b.

Each significant level of the source computed image creates a minimum and maximum in the histogram, which are detected in the next step. Histograms with marked minima resp. the maxima are shown in Figures 7a and 7b.

The minimum and maximum positions in the filtered histogram generate brightness ranges that correspond to objects located at a similar distance from the sensor. Based on these ranges, segments are generated in which elliptical shapes are then detected.

An exception is the first segment that represents the object closest to the sensor in the source data. Experiments have shown that there is no need to segment by maximum for this object, so the first segment is generated only as a region with a brightness range from the beginning to the first minimum of the histogram (the position of the first maximum is deleted from the threshold value list). An example of a segment from such a first brightness range is shown in Figure 7.

Segmented objects are represented by sets of interconnected points (contours) that intersect the ellipse in the next step. Each object can contain several contours and thus several ellipses (Fig. 8). Each contour with its ellipse is subjected to a morphological analysis, where the overlap area is calculated with the inside surface of the ellipse, the area outside the overlap, the occurrence of islands, etc.

-4GB 306898 B6

Based on the morphological analysis and the dimensions of the ellipse, coefficients are determined which are used to remove ellipses that do not correspond to the characteristics of the objects (roundness coefficient, overlap coefficient, occurrence of islands and total ellipse area). An ellipse with parameters that match the set ranges is selected as a potential alarm (alarm) and marked in the source object.

By changing the roundness and overlap coefficient parameters, you can either improve the detection in the disparity map, but at the same time increase the number of false detections in image analysis, or reduce the number of false detections, which in turn worsens the ellipses detection.

Thus, to improve overall detection success, inter-frame consistency analysis, which performs analysis of the current frame, is preferably included in the algorithm against several previous frames. Here he compares already detected ellipses and tries to limit false detection of ellipses. The false detection limit is solved by an algorithm for tracking detected heads in the calculated image (Figs. 9 and 10) and based on several parameters (such as the number of ellipses already detected, distance from the last detection) determines whether the detected ellipse is false detection or detection. head. These detections are stored in the buffer of the track being tracked and can distinguish more persons in the calculated image by evaluating the distance between detected ellipses. Further, in order to evaluate whether the person is moving in the calculated slot, the number of detections in one area in several consecutive calculated pictures is considered. If the detection is inadequate, this is considered a false detection.

When the movement of two persons in the calculated image is detected at the same time, an alarm is set and this type of attack is marked as Piggybacking (Fig. 10) - one authorized person enters the secured area together with another person without authorization.

To detect the second type of attack - Tailgating - when an unauthorized person runs unnoticed after the authorized person, it was necessary to establish a security zone. This zone covers the area just before entering the security area and there are other rules for eliminating false alarms. If one person in the video has been properly detected (at least a few computed images in a row), it is sufficient for the Tailgating alarm to be triggered if the authorized person leaves the security zone (entered the security area) and any movement is detected in the security zone. This security zone is active for a defined period of time to prevent false detections if another person wants to authorize the door after a while (the period is set for the door opening time - external sensor, or several (3 to 5) seconds) .

For the purposes of testing, a database of video sequences with sample and alarm passages was created, the database contained a total of 453 video sequences. The results of testing the proposed algorithm are shown in Table 1.

-5 CZ 306898 B6

Table 1

Screenplay FP FR Total video sequences relative error Model passage ordinary 3 0 187 1.65 Exemplary passage with luggage / tools / tools 10 0 101 9.95 Total exemplary passages 13 0 288 4.55 Tailgating running unauthorized persons 0 4 48 8.3 walking unauthorized persons 0 5 44 11.4 Total Tailgating 0 9 92 9.8% Piggybacking people go close together 0 2 22nd 9.1% people are walking closely behind 0 1 24 4.2% one person crouches next to another / after another 0 8 27 Mar: 29.6% Total Piggybacking 0 11 73 15.1% Total 13 20 May 453 7.3%

PATENT CLAIMS

Claims (9)

PATENT CLAIMS A method of detecting unauthorized passage attempts into protected areas, wherein the area prior to entering the protected area is scanned by a motion sensor adapted to measure the depth map in the camera's field of view, and the depth map images are sent to an evaluation unit, (a) a deep - space reference image of the non - occupied space map is first produced and sent; (b) thereafter, for at least some of the other depth map images in the evaluation unit, a reference image is subtracted therefrom to obtain calculated images, characterized in that the motion sensor is a sensor adapted to measure the depth map in its camera field of view at infrared illumination; that in the evaluation unit prior to step b), reference ranges of the shapes and sizes of the elliptical objects corresponding to the predicted ranges of shapes and sizes of the human heads are stored in the depth map images from the motion sensor, (c) if objects are detected in such calculated images, elliptical objects falling in shape and size within the reference ranges of the elliptical objects and sizes stored in the evaluation unit are detected, the condition being evaluated as an attempted unauthorized passage if either d) a set of consecutive calculated images containing at least one said elliptical shape object followed by a set of at least 1 calculated image without objects, but no more than the number of calculated images without objects corresponding to a maximum of 10 seconds of space capture, immediately followed by at least one calculated image an image with an object and / or -6GB 306898 B6 e) in at least two consecutive calculated images, at least two elliptical objects falling in shape and size within the reference ranges of elliptical objects and sizes stored in the evaluation unit are detected. Method according to claim 1, characterized in that for the detection of objects in step c), d) or e), the images calculated in step b) are segmented using adaptive threshold intensity histograms. Method according to any one of the preceding claims, characterized in that, in order to limit false detection of unauthorized passage attempts, the condition in step e) is evaluated as an unauthorized passage attempt if at least two elliptical-shaped objects falling in shape and size into reference the ranges of shapes and sizes of the elliptical objects stored in the evaluation unit are detected in at least six out of ten consecutive calculated images. Method according to claim 3, characterized in that in step a) the maximum allowable position change of the object with ellipsoidal contours in the immediately consecutive calculated images is stored in the evaluation unit and the state in step is to reduce false detection of unauthorized passage attempts (e) evaluated as an unauthorized passage attempt if the position change of the ellipse-shaped object in pairs of immediately successive calculated images is less than or equal to said maximum permissible position change. Method according to any one of claims 1 to 4, characterized in that in step d) said set of consecutive calculated images without objects corresponds to images taken over a time period of 1 to 10, preferably 3 to 5 seconds. 6. A method of detecting unauthorized passage attempts into protected areas, wherein the area prior to entering the protected area is scanned by a motion sensor adapted to measure the depth map in the camera's field of view, and the depth map images are sent to an evaluation unit, (a) a deep - space reference image of the non - occupied space map is first produced and sent; (b) thereafter, for at least some of the other depth map images in the evaluation unit, a reference image is subtracted therefrom to obtain calculated images, characterized in that the motion sensor is a sensor adapted to measure the depth map in its camera field of view at infrared illumination; that in the evaluation unit, prior to step b), reference ranges of shapes and sizes of elliptical objects corresponding to the predicted ranges of shapes and sizes of human heads in the depth map images from the motion sensor are stored, and the computing unit is connected to the door opening sensor , (c) if objects are detected in such calculated images, elliptical objects falling in shape and size within the reference ranges of the elliptical objects and sizes stored in the evaluation unit are detected, the condition being evaluated as an attempted unauthorized passage if either (d) a set of consecutive calculated images containing at least one said elliptical shape object followed by a set of at least 1 calculated image without objects and immediately followed by at least one calculated image with an object calculated from the image taken at the door opening period; e) in at least two consecutive calculated images, at least two elliptical objects falling in shape and size within the reference ranges of elliptical objects and sizes stored in the evaluation unit are detected. -7 GB 306898 B6 Method according to claim 6, characterized in that for the detection of objects in step c), d) or e), the images calculated in step b) are segmented using adaptive threshold intensity histograms. Method according to claim 6 or 7, characterized in that, in order to reduce false detection of unauthorized passage attempts, the condition in step e) is evaluated as an unauthorized passage attempt if at least two elliptical-shaped objects falling in shape and size in reference the ranges of shapes and sizes of the elliptical objects stored in the evaluation unit are detected in at least six out of ten consecutive calculated images. Method according to claim 8, characterized in that in step a) the maximum allowable position change of the object with elliptical contours in the immediately consecutive calculated images is stored in the evaluating unit and the state in step is to reduce false detection of unauthorized passage attempts. (e) evaluated as an unauthorized passage attempt if the position change of the ellipse-shaped object in pairs of immediately successive calculated images is less than or equal to said maximum permissible position change.