JP2000099686A - Pattern recognition and vehicle recognition method and recording medium storing the program - Google Patents

Abstract

(57) [Summary] [PROBLEMS] To enable a recognition of a specific pattern or a vehicle from image data photographed at a constant moving distance between an observation side and a recognition target. SOLUTION: An observation side on which a line sensor camera is installed and an object including a specific pattern to be recognized move relatively in parallel, and time-series image data obtained by photographing the object at a constant relative movement distance interval. A pattern recognition method for recognizing a specific pattern, comprising: enlarging / reducing a template of a specific pattern recorded in advance in a line axis direction of the line sensor camera, comparing the template with the image data, and Calculate the similarity of
In this method, a specific pattern is recognized from the image data using the similarity of the pattern.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to an image processing method for recognizing a specific pattern or a vehicle from image data captured using a line sensor camera.

[0002]

2. Description of the Related Art There are the following techniques for processing time-series image data photographed outdoors using a line sensor camera. (1) Large image database for railway maintenance management (Reference 1: Proceedings of the Institute of Image Electronics Engineers of Japan 95)
In (2) the road surface inspection device (see Document 2: JP-A-6-129845), the crack detection of the tunnel wall or the rail is performed by a simple filtering process such as an edge extraction and a binarization process. Are doing.

[0003] Also, (3) a method for determining the presence or absence of an object, a method for determining the movement of an object, and an apparatus for implementing the method (Reference 3: Japanese Patent Application No. Hei 9 (1999))
In -317013), an object area is limited from an image captured by a line sensor using a distance sensor,
A method of recognizing a wheel by extracting a circle as a wheel outline in the area is described.

[0004] On the other hand, (4) vehicle extraction is characterized by a front and rear portion of the vehicle for finding a faulty vehicle or a parked vehicle, a method for identifying the vehicle, and a vehicle detection in a parking lot frame for managing a parking lot. It has been widely used for recognition (Reference 4: "Current state and technical problems of image recognition in road traffic systems", IEICE PRMU workshop, 1997)
05/16). These use fixed area sensors and recognize images taken in front and rear of the vehicle in order to recognize many vehicles.

[0005]

However, the observation side such as the line sensor camera and the object to be recognized move relatively in parallel, and the specific pattern is recognized from the time-series image data photographed at a fixed moving distance interval. In consideration of the problem to be solved, conventionally, only simple line figures such as cracks and wheel edges are recognized, and pattern recognition of arbitrary figures is not performed.

Further, in the conventional vehicle recognition technology, no approach has been taken to recognize a vehicle by extracting a specific pattern on the side of the vehicle from an image from a line sensor camera.

[0007] It is an object of the present invention to provide a technique that enables a recognition of a specific pattern or a vehicle from image data photographed at a constant moving distance between an observation side and an object to be recognized.

[0008] The above and other objects and novel features of the present invention will become apparent from the description of the present specification and the accompanying drawings.

[0009]

The following is a brief description of an outline of a representative one of the inventions disclosed by the present application.

(1) A time-series image in which an observation side on which a line sensor camera is installed and an object including a specific pattern to be recognized move relatively in parallel and an object is photographed at a constant relative movement distance interval. A pattern recognition method for recognizing a specific pattern from data, in which a template of a specific pattern recorded in advance in a line axis direction of the line sensor camera is enlarged / reduced, and the template is compared with the image data. A similarity of the pattern is calculated by using the image data, and a specific pattern is recognized from the image data using the similarity of the pattern.

(2) In the pattern recognition method of (1), the step of comparing the image data with the template and calculating the similarity of the pattern is performed by lowering the image resolution of both the image data and the template. This is a procedure for calculating the similarity.

(3) A readable recording medium which records a program for causing a computer to execute the processing procedure of the pattern recognition method of (1).

(4) A readable recording medium which records a program for causing a computer to execute the processing procedure of the pattern recognition method of (2).

(5) The observing side on which the line sensor camera is installed and the vehicle to be recognized move relatively in parallel, and the stopped vehicle is determined from time-series image data obtained by photographing the vehicle at a constant relative moving distance interval. A vehicle recognition method for recognizing, wherein a template of a specific pattern recorded in advance in a line axis direction of the line sensor camera is enlarged / reduced, and the template is compared with the image data to determine a similarity of a wheel pattern. This is a method of calculating a degree, associating the same wheel based on the similarity of the wheel pattern, and recognizing a vehicle including the associated wheel set.

(6) In the vehicle recognition method of (5), the step of comparing the image data with the template and calculating the similarity is performed by lowering the image resolution of both the image data and the template and comparing the image data and the template. This is the procedure for calculating the degree.

(7) A readable recording medium which records a program for causing a computer to execute the processing procedure of the vehicle recognition method of (5).

(8) A readable recording medium which records a program for causing a computer to execute the processing procedure of the vehicle recognition method of (5).

That is, according to the present invention, the observing side where the line sensor camera is installed and the object including the specific pattern to be recognized move relatively in parallel, and the side of the object is photographed at a constant relative moving distance interval. A method for recognizing a specific pattern from the time-series image data obtained, wherein a procedure for enlarging / reducing a template of a specific pattern recorded in advance in the line axis direction of the line sensor camera is provided. And calculating the similarity, and recognizing a specific pattern from the image data using the similarity.

According to the present invention, the specific pattern and the vehicle to be recognized can be recognized from the time-series image data taken at a constant relative movement distance when the observation side and the recognition target move relatively in parallel. It becomes possible. Further, it is possible to realize a pattern recognition device that can be moved indoors and outdoors, instead of the conventional fixed pattern recognition device having a limited installation place.

[0020]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The following embodiments (examples) of the present invention will be described with reference to the drawings.

(Embodiment 1) FIG. 1 is a block diagram showing a schematic configuration of a pattern recognition apparatus according to an embodiment (embodiment) 1 of the present invention. As shown in FIG. 1, a pattern recognition device according to the first embodiment (embodiment) includes an information processing device (CP).
A storage device 102 and a display device 103 such as a display are connected to the U) 101 via a bus line 109.

FIG. 2 is a flowchart showing a processing procedure of the specific pattern recognition method according to the first embodiment of the present invention. In the specific pattern recognition method according to the first embodiment, as shown in FIG. 2, the observation side on which the line sensor camera is installed and the object including the specific pattern to be recognized relatively move in parallel, and Time-series image data S101 of an object photographed at the relative moving distance interval is stored in the storage device 102. The read time-series image data (S101) is input to the information processing apparatus (CPU) 101, and the input time-series image data (S101) is subjected to image binarization and wheel pattern erasing. Processing is performed (input processing: S102). The template image (S103) of the specific pattern stored in the storage device 102 in advance is
The image data is enlarged / reduced in the line axis direction of the line sensor camera, and the pre-processed image data is compared with a template to calculate a pattern similarity (S105). A specific pattern is recognized from the image data based on the similarity of the pattern (S106). The recognized specific pattern is displayed on the display device 103 (S107). The details of the method of calculating the similarity of the pattern are described in Japanese Patent Application No. 9-317013 of Document 3. Here, it is omitted.

According to the first embodiment, the observation side and the object to be recognized move relatively in parallel, and the specific pattern to be recognized is recognized from the time-series image data photographed at a fixed relative movement distance interval. It becomes possible. Also, instead of the conventional fixed pattern recognition device where the installation location is limited, indoors,
It is possible to realize a pattern recognition device that can be moved regardless of the outdoors.

(Embodiment 2) In this embodiment (embodiment 2), the observation side of the present invention is a moving vehicle, and the specific pattern to be recognized is the wheels of a stopped vehicle. The time-series data image is set on the side of the moving vehicle and is an image of the side of the stopped vehicle. Here, when the line sensor camera is not installed perpendicularly to the moving direction, the image becomes longer while the front or rear part of the stopped vehicle is captured, but it does not matter.

FIG. 3 is a block diagram showing a schematic configuration of the vehicle recognition apparatus according to the second embodiment of the present invention. A storage device 102, a display device 103 such as a display, and a measurement device 104 are connected to an information processing device (CPU) 101 illustrated in FIG. However, in the storage device 102, a measured value storage unit 107 and an object shape feature storage unit 108 are provided. In the measuring device 104, a distance sensor 105 and a line sensor camera 106 are provided. Here, the distance sensor 105
Is a distance measuring device such as a millimeter-wave radar or an ultrasonic sensor.

FIG. 4 shows an apparatus for acquiring image data in the vehicle recognition method according to the second embodiment. In FIG. 4, a vehicle including a wheel, which is a specific pattern to be recognized, is a stopped vehicle 201 on a road shoulder, and an observation side including a line sensor camera 205 is a moving vehicle 202. Line sensor camera 2
The line axis 05 is set perpendicular to the ground.
The moving vehicle 202 is provided with a rotary encoder 204 for recording the output from the line sensor camera 205 at a constant traveling distance of the moving vehicle 202.

The rotary encoder 204 is a measuring instrument used to detect the angular change and angular velocity of the part of the mechanical device that makes a rotational movement, and can measure the moving distance from the circumference of the rotary and the number of rotations. In the present invention, the image data obtained by simply synchronizing the output from the licensor camera with the measured moving distance has a real size in the moving direction.

FIG. 5 is a diagram showing a line sensor image and a wheel template image according to the second embodiment. Reference numeral 301 denotes a line sensor image, 302 denotes a wheel, 401 denotes a wheel image, and 40 denotes a wheel image.
Reference numeral 2 denotes a wheel image after the hole filling process, and reference numeral 501 denotes a wheel template image. FIG. 6 is a diagram showing an image of a wheel area in the recognized image data of the second embodiment. ).

As shown in FIG. 5, the wheel 302 is composed of a tire portion and a wheel portion. Generally, in a grayscale image, the tire portion is darker than the wheel portion. Therefore, binarization processing (Reference 5: “Image Analysis Handbook”, published by the University of Tokyo) is performed so as to provide a threshold value between the luminance value of the tire portion and the luminance value of the wheel portion. Separate like black. Line sensor image 301
Since the wheel pattern remains in the wheel portion in the image obtained by binarizing the, the morphology (Reference 6: “Morphology”, Corona Corporation) is used to fill in the holes. Therefore, the wheel image 401 becomes the wheel image 402 after the hole filling process in which the wheel pattern disappears ((b) of FIG. 5). The wheel template image 501 shown in FIG. 5C is recorded (stored) in a storage device such as a memory in advance.

FIG. 7 is a flowchart showing a processing procedure of a wheel recognition method according to the pattern recognition method of the second embodiment.

In the wheel recognition method according to the second embodiment, as shown in FIG. 7, the moving vehicle 202 on the observation side where the line sensor camera 205 is installed and the stopped vehicle 201 to be recognized move relatively in parallel. Then, the time-series image data S <b> 201 of the stopped vehicle 201 obtained by photographing the stopped vehicle 201 at a constant relative moving distance interval is stored in the storage device 102. The read time-series image data (S201) is input to the information processing device (CPU) 101, and the input time-series image data (S201) is subjected to image binarization and wheel pattern erasure, etc. Processing is performed (S202), and the data is divided into blocks (S203). The time-series image data divided into the blocks is processed by an information processing device (CPU) 101.
On the other hand, a wheel template image (template image of a specific pattern) 501 (S204) is input.

In the information processing device (CPU) 101, enlargement / reduction processing of the template image 501 in the line axis direction is executed (S205), and one block in the time-series image data and the wheel template image 501 (S20) are processed.
The similarity is calculated by calculating a correlation value and a difference value with the image obtained by enlarging / reducing 4) (S206). It is checked whether the calculated similarity is equal to or greater than a certain threshold (S20).
7). If the similarity is equal to or greater than a certain threshold, the block of the image data is recognized as having wheels (S208), and the result is displayed on the display device 103 (S211). If the similarity is equal to or less than a certain threshold value, it is checked whether or not the magnification of the enlargement / reduction processing of the wheel template image has been changed within a certain range (S209). The block recognizes that there is no wheel (S210), and displays the result on the display device 103 (S211).

Next, when the pattern recognition processing for all blocks of the image data is completed, the pattern recognition processing is completed. When the pattern recognition processing for all blocks is not completed, the processing returns to step (S205) and the above processing is repeated. (S2
12).

In the similarity calculation, the difference in wheel size for each vehicle can be rounded to the range of the quantization error by reducing the image resolution of the time-series image data and the wheel template image. Therefore, it is possible to calculate the similarity between wheels of a plurality of sizes using one wheel template image.

FIG. 8 is a flowchart showing a processing procedure of a method for recognizing a vehicle using the wheel recognition method of the second embodiment.

In the vehicle recognition method according to the second embodiment, as shown in FIG. 8, wheels are recognized using the wheel recognition method (S301). Next, it is checked whether the vehicle is stopped (S302). If the vehicle is stopped, the wheels of the same vehicle are associated with the recognized wheel set based on the similarity of the wheel pattern. (In calculating the similarity, the wheel image is obtained from the scaling process in the line axis direction. The process is performed by returning the aspect ratio to the aspect ratio of 1: 1) (S303). (The details of the calculation of the similarity are described in the specification of Japanese Patent Application No. 9-317013).

That is, when a stopped vehicle is parked in parallel,
Since the inter-vehicle distance may be shorter than the inter-wheel distance 601 (FIG. 6), in S303, the image similarity of the recognized wheel set is calculated, and the wheels are associated based on the similarity. Recognize. This similarity can be obtained as follows. The images of the wheel area in the image data recognized in the second embodiment are assumed to be a wheel image A 602 and a wheel image B 603 as shown in FIG. At the time of wheel recognition, the aspect ratio of the image data is obtained in the enlargement / reduction processing in the line axis direction. Therefore, based on this ratio, the wheel image is returned to the actual 1: 1 aspect ratio, and the image similarity is calculated. However, there is a case where the similarity is calculated by rotating one wheel image. In this way, it is possible to recognize a stopped vehicle.

If the vehicle is not stopped, the recognized wheel set is associated with the wheels of the same vehicle by thresholding the inter-wheel distance (S304). The vehicle including the associated wheel set is recognized (S305).

That is, when the vehicle to be recognized is moving, the distance between adjacent wheels of the same vehicle (inter-wheel distance 601) is generally smaller than the inter-vehicle distance. Therefore, a wheel set including the same vehicle is determined by recognizing the wheels and performing threshold processing on the inter-wheel distance 601 as in the second embodiment. With this method, it is possible to recognize the moving vehicle.

As can be seen from the above description, according to the second embodiment, by using the vehicle recognition method, the observation side and the object to be recognized move relatively, and the vehicle is recognized from the captured image. can do.

As described above, the present invention has been specifically described based on the above embodiments (examples). However, the present invention is not limited to the above embodiments and can be variously modified without departing from the gist thereof. Needless to say.

[0042]

As described above, according to the present invention,
By using the pattern recognition method or the vehicle recognition method, a specific pattern or a vehicle can be recognized from an image captured by moving the observation side and the object to be recognized relatively.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a schematic configuration of a pattern recognition apparatus according to Embodiment 1 of the present invention.

FIG. 2 is a flowchart illustrating a processing procedure of a specific pattern recognition method according to the first embodiment of the present invention.

FIG. 3 is a block diagram illustrating a schematic configuration of a vehicle recognition device according to a second embodiment of the present invention.

FIG. 4 is an apparatus for acquiring image data in the vehicle recognition method according to the second embodiment.

FIG. 5 is a diagram illustrating a line sensor image and a wheel template image according to the second embodiment.

FIG. 6 is a diagram showing an image of a wheel area in recognized image data according to the second embodiment.

FIG. 7 is a flowchart illustrating a processing procedure of a wheel recognition method according to the pattern recognition method of the second embodiment.

FIG. 8 is a flowchart illustrating a processing procedure of a method for recognizing a vehicle using the wheel recognition method according to the second embodiment.

[Explanation of symbols]

101: information processing device (CPU), 102: storage device,
103: display device, 104: measuring device, 105: distance sensor, 106: line sensor camera, 107: measured value accumulating unit, 108: object shape characteristic accumulating unit, 109: bus line, 201: stopped vehicle, 202: moving vehicle , 203 ... ultrasonic sensor, 204 ... rotary encoder, 205 ...
Line sensor camera, 301: Line sensor camera image, 301A: Wheel, 302: Wheel image, 303: Wheel image after filling processing (wheel template image), 601
... wheel distance, 602 ... wheel image A (wheel texture A), 603 ... wheel image B (wheel texture B).

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Akio Shio 3-19-2 Nishishinjuku, Shinjuku-ku, Tokyo Inside Japan Telegraph and Telephone Corporation (72) Inventor Sakuichi Otsuka 3-19 Nishishinjuku, Shinjuku-ku, Tokyo No. 2 Nippon Telegraph and Telephone Corporation F term (reference) 5B057 AA16 CA08 CA12 CA16 CC01 CD05 DA06 DC30 DC33 5H180 AA01 CC04 CC11 LL01 LL04 5L096 BA04 CA02 DA02 EA03 EA45 FA34 FA39 HA08 JA09

Claims (8)

[Claims] 1. Time-series image data in which an observation side on which a line sensor camera is installed and an object including a specific pattern to be recognized move relatively in parallel and an object is photographed at a constant relative movement distance interval. A pattern recognition method for recognizing a specific pattern from a pattern sensor, wherein a template of a specific pattern recorded in advance in a line axis direction of the line sensor camera is enlarged / reduced, and the template is compared with the image data. A pattern recognition method comprising: calculating a pattern similarity; and recognizing a specific pattern from the image data using the pattern similarity. 2. The pattern recognition method according to claim 1, wherein the step of comparing the image data with the template and calculating the similarity of the pattern includes reducing the image resolution of both the image data and the template. A pattern recognition method, which is a procedure for calculating a degree of similarity. 3. A procedure in which an observation side on which a line sensor camera is installed and an object including a specific pattern to be recognized move relatively in parallel, and a time series in which the object is photographed at a constant relative movement distance interval. A procedure for obtaining image data; a procedure for enlarging / reducing a template of a specific pattern recorded in advance in the line axis direction of the line sensor camera; A readable recording medium on which a program for causing a computer to execute a procedure for calculating a degree is recorded. 4. The recording medium according to claim 3, wherein the step of comparing the image data and the template to calculate the similarity of the pattern includes reducing the image resolution of both the image data and the template and comparing the image data and the template. A recording medium characterized by a procedure for calculating a degree. 5. A stopped vehicle is recognized from time-series image data obtained by photographing a vehicle at a fixed relative movement distance, wherein an observation side on which a line sensor camera is installed and a vehicle to be recognized move relatively in parallel. A vehicle recognition method, comprising: expanding and reducing a template of a specific pattern recorded in advance in a line axis direction of the line sensor camera; comparing the template with the image data to determine a similarity between wheel patterns; And a vehicle recognizing method for recognizing a vehicle including the wheel set associated with the same wheel based on the similarity of the wheel patterns. 6. The vehicle recognition method according to claim 5, wherein the step of comparing the image data and the template to calculate the similarity includes reducing the image resolution of both the image data and the template and comparing the image data and the template. A vehicle recognition method comprising calculating a degree. 7. A procedure in which an observing side on which a line sensor camera is installed and a vehicle to be recognized move relatively in parallel, and a time-series image data of the vehicle taken at a fixed relative moving distance interval is used to detect the vehicle. A procedure for recognizing a wheel pattern, and enlarging and pre-recording a template of a specific pattern recorded in advance in the line axis direction of the line sensor camera.
A step of reducing, a step of comparing the image data and the template to calculate a similarity of a wheel pattern, a step of associating the same wheel based on the similarity of the wheel pattern, A readable recording medium storing a program for causing a computer to execute a procedure for recognizing a vehicle including a wheel set. 8. The recording medium according to claim 7, wherein the step of comparing the image data and the template to calculate the similarity includes reducing the image resolution of both the image data and the template and comparing the wheel patterns. A recording medium, which is a procedure for calculating the similarity of a recording medium.