JP2013045188A - Object detection method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To solve the problems that individual size becomes small and it cannot be recognized that a car exists by a fact that one object is detected by being separated into a plurality of objects depending on a binarization threshold since the binarization threshold becomes higher or lower according to environmental change when a fact that the car exists in a monitoring area is recognized by difference, binarization, noise removal, labeling, and size determination processing under environment that both of a person and the car enter the monitoring area.SOLUTION: Authentication accuracy of an existence area of an object is raised by using normalization correlation. As a result, performance for recognizing that a car exists is raised by stopping processing for prevention of erroneous detection when the car exists in the monitoring area, and combining the normalization correlation processing and differential processing when an intruder is detected when the car does not exist.

Description

  The present invention relates to a technique for detecting an intruder or an intruder in a CCTV (Closed Circuit Tele-Vision) device for monitoring an intruder or an intruder.

  In the image processing apparatus that detects a person who enters the monitoring area, an operation of stopping the process when a car enters will be described with reference to FIG. FIG. 4 is a flowchart for explaining a conventional object detection method.

  In FIG. 4, first, in steps 401 to 403, the background image and the threshold image are initialized.

In an input image capture step 401, an input image used for background image creation, threshold creation, and object detection is captured from the camera.
In the background image / threshold creation step 402, the acquired input image is weighted and averaged, and used to create and detect a reference background image having no object to be detected such as a person or a vehicle in the monitoring area. A threshold value is created (for example, refer to Patent Document 1).
In the background image creation end determination step 403, the number of frame processes is determined, and it is determined whether or not the number of frame processes is equal to or greater than a specified value. If the number of frame processing is less than the specified value, the process returns to the input image capturing step 401 and the processing of steps 401 to 403 is repeated. If the number of frame processing is greater than or equal to the specified value, the input image capturing step The process proceeds to 404.

  Object detection processing includes input image capture step 404, difference processing step 405, binarization processing step 406, noise removal processing step 407, labeling processing step 408, size detection processing stop determination step 409, object determination processing step. 410, an intruder determination step 411 performs an object detection process, an alarm process step 412 performs an alarm process, and an image display process step 414 performs an image display process. In addition, a background image / threshold creation step 413 is performed between the alarm processing step 412 and the video display processing step 414.

In the input image capture step 404, as in step 401, a video is captured from the camera and an input image is created.
In the difference processing step 405, the difference between the captured input image and the latest background image created in the background image / threshold creation step 402 or 413 is calculated, and a difference image is created. Note that the difference process may be performed with a positive or negative sign or an absolute value for the calculation of the input image-background image.
In the binarization processing step 406, the difference image obtained in the difference processing step 405 is obtained by multiplying the latest threshold value obtained in the background image / threshold value creation processing step (step 402 or step 413) by a coefficient K. Binarization is performed (K is an integer of 1 to 255, for example).
The binary image obtained in the binarization processing step includes noise composed of several pixels. For this reason, in the noise removal processing step 407, noise composed of several pixels is removed from the binary image to create a noise-removed image.
In the labeling processing step 408, labeling, width, height, and area calculation are performed for each clump for the noise-removed image obtained in the noise removing step 407.

In the process stop determination step 409 based on the size, it is determined whether or not there is an object corresponding to the size of the vehicle using the size of the object determined from the width and height of each block obtained in the labeling process step 408. Determine. If there is a vehicle, the process proceeds to the background image / threshold creation step 413 without performing the determination process. If there is no vehicle, the process proceeds to the object determination processing step 410.
In the object determination processing step 410, it is determined whether or not the person is an intruder for each cluster obtained in the labeling processing step 408 using the size (width and height) of the object, the staying time, the moving distance, and the like.
In the intruder determination step 411, it is determined whether or not there is an object determined as an intruder in the object determination processing step 410. If there is an intruder, the process proceeds to the alarm processing step 412, and if there is no intruder, the process proceeds to the background image / threshold creation processing step 413.
In an alarm processing step 412, an alarm is output to an external device, such as turning on a rotating lamp or a laminated display lamp. Other alarm outputs include sounding a buzzer and notifying the recording device, and at least one of them is executed.

Next, in the background image / threshold value creation processing step 413, every time an input image is captured, the same as the background image / threshold value creation step 402 is performed in order to follow the video state that changes every moment in response to a change in the shooting environment. Processing, that is, creation of a reference background image and creation of a threshold used for detection.
In the video display processing step 414, when there is an intruder, the characters “intruder detection” and a marker indicating the detected portion are superimposed on the input video, and when there is no intruder, an input image is output and monitored. Displayed on the monitor monitored by the employee.
After the video display processing step 414, the process proceeds to the input image capturing step 404 again, and the object detection processing is continued.

JP 2009-282975 A

The weak point of the vehicle determination in the above-described prior art will be described with reference to FIGS. FIG. 5 is a schematic diagram for explaining an example of a conventional object detection method. Reference numeral 501 denotes an input image, and reference numerals 502 to 505 denote objects such as an intruder or an intruding object detected from the input image 501. The object 502 is an intruder, and the objects 503 to 505 are part of an intruding object (for example, an intruding vehicle).
In the conventional technique described above, binarization processing is performed for each pixel by a value obtained by multiplying the threshold value obtained in the background image / threshold value creation processing step 402 by a coefficient. For this reason, for example, there is a pixel having a large threshold value for a certain period of time due to an increase or disappearance of a shadowed portion due to the influence of sunshine change or blinking of illumination. In such a case, for example, a person can recognize the object 502 as one object as in the input image 501 shown in FIG. However, the car may be detected by being divided into objects 503 to 505.
In the case of FIG. 5, in order to determine whether or not each of the objects 503 to 505 is a car size, the height is calculated to be smaller than the car size, and it cannot be recognized that a car exists.
In the conventional object detection method, since the binarization threshold is determined for each pixel using a weighted average, when a change in sunshine occurs, the binarization threshold is set to a relatively large value for a certain period of time, The probability of being divided and detected increases. When an object is detected in a plurality of parts, the size of each object is small, so it may be determined that there is no car even though the car exists.

As described above, when the existence of a vehicle in the monitoring area is conventionally performed by difference, binarization, noise removal, labeling, and size determination processing, the threshold for binarization changes to the environment. Depending on the binarization threshold, one object is detected in multiple parts, so that the individual size is reduced and there is a car. I could not.
In view of the above problems, an object of the present invention is to provide an object detection method with improved object detection accuracy.

  In order to achieve the above object, an object detection method of the present invention takes an input image from a camera that images a monitoring area, performs image processing on the acquired input image, performs object detection processing, and intrudes into the monitoring area. In the object detection method of the object detection device for detecting a person or an intruder, the input image is divided into a plurality of blocks, an existence region of the object is determined from a correlation value and a difference value for each of the divided blocks, and the object When the presence area is equal to or greater than a predetermined threshold, the object detection process is stopped.

  According to the present invention, it is possible to improve the recognition accuracy of the object existence region by using the normalized correlation. In addition, since a certain size and pattern are necessary for recognition by normalized correlation, the object existence area is determined not for each pixel but for each block, and the object existence area is difficult to be divided. .

It is a flowchart for demonstrating the detection stop determination process step in one Example of the object detection method of this invention. It is a schematic diagram for demonstrating one Example of the object detection method of this invention. It is a flowchart for demonstrating one Example of the object detection method of this invention. It is a flowchart for demonstrating the conventional object detection method. It is a schematic diagram for demonstrating an example of the conventional object detection method. It is a figure which shows the structure of the monitoring apparatus which concerns on this invention.

In an environment where both people and vehicles enter the monitoring area, the present invention stops processing to prevent false detection when a vehicle exists in the monitoring area, and detects an intruder when there is no vehicle. In this case, the performance of recognizing the presence of a vehicle is improved by combining the normalized correlation process and the difference process.
In the above processing, the recognition accuracy of the object existing area is improved by using the normalized correlation. In addition, since normalized correlation may be low in areas with uniform patterns (brightness) such as asphalt, in addition to normalized correlation, differences are also combined, normalized correlation is less than a certain level, and presence area detection A portion where a difference greater than or equal to a threshold value (a value smaller than object detection) occurs is set as an object existence region, thereby compensating for a reduction in recognition accuracy in the normalized correlation.
Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. In addition, in the description of each drawing including FIG. 4 and FIG. 5 which have already been described in the prior art, the same reference numerals are assigned to the components having a common function, and the description is avoided as much as possible.
Moreover, the following description is for describing one embodiment of the present invention, and does not limit the scope of the present invention. Accordingly, those skilled in the art can employ embodiments in which these elements or all of the elements are replaced with equivalent ones, and these embodiments are also included in the scope of the present invention.

FIG. 6 is a diagram showing a configuration of a monitoring CCTV apparatus according to the present invention. The surveillance CCTV device includes a camera 601, an object detection device 600, an instruction device 609, and a display device 610.
The object detection apparatus 600 includes a video input circuit 602, an image processor 603, a program memory 604, a work memory 605, an external I / F circuit 606, a video output circuit 607, and a bus 608. . The video input circuit 602, the image processor 603, the program memory 604, the work memory 605, the external I / F circuit 606, and the video output circuit 607 are connected to the bus 608. A bus 608 is a data bus for each device in the object detection apparatus 600 to access each other.

The camera 601 captures an area to be monitored (monitoring area), and outputs the captured image to the object detection apparatus 600 as an input image.
In the object detection device 600, an image input from the camera 601 is input to the video input circuit 602. The video input circuit 602 takes an input image and outputs it to the work memory 605. The work memory 605 records an input image.
The image processor 603 processes the input image recorded in the work memory 605 according to the program recorded in the program memory 604, and outputs the processing result to the display device 610 via the video output circuit 607.
The display device 610 displays the processing result input from the video output circuit 607 on the display unit.
The instruction device 609 is an input device for an operator to operate the object detection device 600 using an instruction device 609 such as a mouse or a keyboard. The image processor 603 changes the program parameters stored in the program memory 604 based on an instruction from the operator input via the external I / F circuit 606 from the instruction device 609 while changing and correcting the parameters of the program. Process the image.

An embodiment of the object detection method of the present invention will be described with reference to FIG. FIG. 3 is a flowchart for explaining an embodiment of the object detection method of the present invention. 3 is executed by the image processing processor 603 of the object detection apparatus 600 described with reference to FIG. 6 in accordance with a program recorded in the program memory 604.
3 deletes the process stop determination step 409 depending on the size from the flowchart of the conventional object detection method described in FIG. 4, and before the difference process step 405, the detection stop determination process step 301 and the detection process. An execution determination step 302 is added. Details of the detection stop determination processing step 301 will be described later.
In the detection process execution determination step 302, only when it is determined as “detection process execution” based on the result (“detection process execution” or “detection process stop”) determined in the detection stop determination process step 301, Difference processing step 405, binarization processing step 406, noise removal processing step 407, labeling processing step 408, object determination processing step 410, intruder determination step 411, warning processing step 412, background image / threshold generation processing step 413, and A video display processing step 414 is performed.
If the detection stop determination processing step 301 determines “detection processing stop”, the process returns to the input image capture step 404 to capture the next input image and perform object detection processing.

Next, details of the detection stop determination processing step 301 will be described with reference to FIGS. 1 and 2. FIG. 1 is a flowchart for explaining detection stop determination processing steps in an embodiment of the object detection method of the present invention. FIG. 2 is a schematic diagram for explaining an embodiment of the object detection method of the present invention. 2A shows the background image 201 and FIG. 2B shows the input image 202. FIG. 2A and FIG. 2B both illustrate processing in which the image width is divided into (M + 1) blocks and the image height is divided into (N + 1) blocks. is there. B _{r00 to} B _rM0 are the _{0th to Mth} blocks in the first row of the background image 201, and _{Br0N to BrMN} are the _{0th to Mth} blocks in the Nth row of the background image 201. B _{n00 to} B _nM0 are the _{0th to Mth} blocks in the first row of the input image 202, and _{Bn0N to BnMN} are the _{0th to Mth} blocks in the Nth row of the input image 202. Further, in the background image 201 and the input image 202, the width of one block is W _B and the height is H _B.
Furthermore, in the background image 201 and the input image 202 in FIG. 2, a person and a vehicle passing through the gate are monitored. Therefore, the background image 201 shows an image with a gate and a fence, and the input image 202 shows an object 203 indicating a person and an object 204 indicating a car in addition to the gate and the fence.
The detection stop determination processing step 301 divides the background image and the input image into (M + 1) blocks and the image height (N + 1) blocks, and normalization correlation of each divided block. This is a process for determining an object presence area based on a value and a difference value.
1 is also executed by the image processing processor 603 of the object detection apparatus 600 described in FIG. 6 according to a program recorded in the program memory 604.

In FIG. 1, in the object existence area number initialization step 101, the object existence area counter C is initialized to 0 (C = 0).
In the processing count N determination step 102, it is determined whether or not the block processing count Pn in the height direction of the image is N or less. If the processing number Pn is N or less (Pn ≦ N), the process proceeds to the processing number M determination step 103. If the processing number Pn is greater than N (Pn> N), the object presence region determination processing step 108 is performed. Migrate to After the determination, 1 is added to Pn and the process proceeds to the next process (Pn = Pn + 1).
In the process count M determination step 103, it is determined whether or not the block process count Pm in the width direction of the image is M or less. If the number of processing times Pm is N or less (Pm ≦ M), the process proceeds to the process of calculating the average correlation value 104. If the number of processing times Pm is greater than M (Pn> N), the process of the processing number N determination step 102 Transition. After the determination, 1 is added to Pm and the process proceeds to the next process (Pm = Pm + 1).

In the average correlation value calculation step 104, when the number of processing times Pn = 0 and the number of processing times Pm = 0, the normalized correlation between Br _{00 of the} background image 201 and Bn ₀₀ of the input image 202 is calculated. By using the normalized correlation, even if a sunshine change or the like occurs rather than the difference and the level changes, it is difficult to be affected. Therefore, the recognition accuracy of the object existing area is higher than the difference processing.
In the average difference value calculation step 105, the difference (input image-background image, with a positive or negative sign, or absolute value) is calculated for each pixel of Br _{00 of the} background image 201 and Bn ₀₀ of the input image 202. Either may be calculated) and an average difference value is calculated.
In the normalized correlation, since the correlation value cannot be obtained correctly in the portion without the pattern, it is determined that there is no object in the portion having a low average difference value even if the normalized correlation value is less than 0.95.

In the object region determination step 106 using the correlation value and the difference, the average correlation value Ra obtained in the average correlation value calculation step 104 is compared with a predetermined threshold value Thc and obtained in the average difference value calculation step 105. The average difference value Da is compared with a predetermined threshold Ths, and it is determined whether or not the object is in an area.
That is, if Ra <Thc and Da> Ths, it is determined that an object is present, and the process proceeds to the object presence area increment step 107. Otherwise, the process returns to the process M determination step 103.
In the object presence area increment step 107, the object presence area is counted up. In this embodiment, Thc = 0.95 (the object exists when it differs from the background by 5% or more), and Ths = 9 (the object exists when the difference is 9 or more). For example, when the processing count Pn = 0 and the processing count Pm = 0 in the average correlation value calculation step 104, the processing count Pn = 0 and the processing count Pm = 1 are obtained as a result of the processing in step 107, and Br of the background image 201 is obtained. ₁₀ and Bn _{10 of the} input image 202, it is determined whether or not an object exists in an average correlation value calculation step 104, an average difference value calculation step 105, and an object region determination step 106 using the correlation value and the difference. If it is determined that an object exists, the object presence area is incremented in an object presence area increment step 107.
This process is repeated until the number of processes Pn = N and the number of processes Pm = M. When the processing number Pn = N and the processing number Pm = M, the process proceeds to the detection process stop determination step 108.

In the detection process stop determination step 108, “object existence area> ThA”
When is established, it is determined that a vehicle is present, and the process proceeds to the detection process stop flag setting step 110. If “object existence area> ThA” is not established, the process proceeds to the detection process execution flag setting step 109.
In a detection process execution flag setting step 109, a detection process execution flag is set.
In the detection process stop flag setting step 110, a detection process stop flag is set.

According to the embodiment of FIGS. 1 to 3, the recognition accuracy of the existence region of the object is improved by using the normalized correlation. In addition to the normalized correlation, the combination of the differences can also prevent the recognition accuracy from being lowered at an inner place where the pattern is small (brightness is uniform). Further, since the object existence area is determined for each block, not for each pixel, the object existence area is difficult to be divided.
As a result, in an environment where both people and vehicles enter the monitoring area, when there is a vehicle in the monitoring area, processing is stopped to prevent false detection, and intruder detection is performed when there is no vehicle. The performance (recognition accuracy) for recognizing the presence of a vehicle is improved by combining the normalized correlation processing and the difference processing.

  101: Object existence region number initialization step, 102: Processing number N determination step, 103: Processing number M determination step, 104: Average correlation value calculation step, 105: Average difference value calculation step, 106: Object region determination step : Object presence region increment step, 108: detection process stop determination step, 109: detection process execution flag setting step, 110: detection process stop flag setting step, 201: background image, 202: input image, 203, 204: object, 301 , Detection stop determination processing step, 302: detection processing execution determination step, 401: input image capture step, 402: background image / threshold creation step, 403: background image creation end determination step, 404: input image capture step, 405 : Difference processing step, 406: Binarization processing step, 407: Noise removal processing step, 408: Labeling processing step, 409: Detection processing stop determination step based on size, 410: Object determination processing step, 411: Intruder determination step, 412: Alarm processing Step 413: Background image / threshold creation processing step 414: Video display processing step 414, 501: Input image, 502-505: Object, 600: Object detection device, 601: Camera, 602: Video input circuit, 603: Image Processor 604: Program memory 605: Work memory 606: External I / F circuit 607: Video output circuit 608: Communication path 609: Instruction device 610: Display device

Claims (1)

  In the object detection method of the object detection device for capturing an input image from a camera that captures a monitoring area, subjecting the captured input image to image detection processing, and detecting an intruder or intruder in the monitoring area, The input image is divided into a plurality of blocks, an object existing area is determined from a correlation value and a difference value for each of the divided blocks, and the object detection process is performed when the object existing area is equal to or greater than a predetermined threshold. The object detection method characterized by stopping.

Images