WO2015098527A1

WO2015098527A1 - Detection device, method for detecting object to be detected, and control program

Info

Publication number: WO2015098527A1
Application number: PCT/JP2014/082712
Authority: WO
Inventors: 健太西行; 健太永峰
Original assignee: 株式会社メガチップス
Priority date: 2013-12-27
Filing date: 2014-12-10
Publication date: 2015-07-02
Also published as: JP6378483B2; JP2015125696A

Abstract

A storage unit stores a background model which includes background image information. Using the background model and an inputted image, a detection unit detects an object to be detected which is present in a captured image region which appears in the inputted image. On the basis of the result of the detection by the detection unit, a specification unit specifies a region in the captured image region where the probability is high that the object to be detected is present. Using the result of the specification with the specification unit, a determination unit determines whether to register the image information obtained from the inputted image into the background model as background image information.

Description

Detection apparatus, detection object detection method, and control program

The present invention relates to a detection technique for a detection target.

As described in Patent Document 1 and Non-Patent Documents 1 to 3, various techniques have been proposed for detecting a detection object such as a person.

JP 2012-155717 A

When detecting a detection target, it is desired to improve its accuracy.

Therefore, the present invention has been made in view of the above-described points, and an object thereof is to provide a technique capable of improving the detection accuracy of a detection object.

One aspect of the detection apparatus according to the present invention uses a storage unit that stores a background model including background image information, and the background model and the input image to detect a detection target existing in an imaging region that appears in the input image. A detection unit for detecting, a specifying unit for specifying a region where a detection target object is likely to exist in the imaging region based on a detection result of the detection unit, and image information obtained from the input image as a background image A determination unit that determines whether or not to register the information in the background model as a result using a specific result of the specifying unit.

In the detection device according to the aspect of the invention, the specifying unit may detect the detection target object for each of a plurality of partial imaging regions constituting the imaging region based on a detection result of the detection unit. There is a possibility that the detection target is present in a partial imaging area in which the detection frequency is greater than or equal to the first threshold value or greater than the first threshold value among the plurality of partial imaging areas. High area.

Moreover, in one aspect of the detection apparatus according to the present invention, the specifying unit obtains the non-detection frequency of the detection object for each of the plurality of partial imaging regions based on the detection result of the detection unit, The specifying unit includes the detection frequency and the non-detection of a partial imaging region in which the non-detection frequency is greater than or equal to the second threshold value or greater than the second threshold value among the plurality of partial imaging regions. Clear the frequency.

In the detection device according to the aspect of the invention, the specifying unit may increase the detection frequency of the first partial imaging region where the detection target is detected in the plurality of partial imaging regions, and The detection frequency of the second partial imaging area around the one partial imaging area is increased.

Further, one aspect of the detection target object detection method according to the present invention is a detection target object detection method in a detection device, wherein (a) a background model including background image information and an input image are used. A step of detecting a detection target existing in an imaging region shown in the input image; and (b) a region where the detection target is likely to exist in the imaging region based on the detection result in the step (a). And (c) determining whether or not to register image information obtained from the input image as background image information in the background model using the identification result in the step (b). .

Moreover, one aspect of the control program according to the present invention is a control program for controlling a detection device that detects a detection object, and (a) a background model including background image information and an input to the detection device. A step of detecting a detection object existing in an imaging region shown in the input image using an image; and (b) a region in the imaging region where the detection target is highly likely to exist ) Based on the detection result in step (b), and (c) whether the image information obtained from the input image is registered in the background model as background image information is determined using the result specified in step (b). And a step of determining the process.

Detecting object detection accuracy is improved.

It is a figure which shows the outline | summary of the process performed with a detection apparatus. It is a figure which shows the structure of a detection apparatus. It is a figure which shows the structure of an image process part. It is a figure which shows an example of an imaging area. It is a figure which shows an example of a background model. It is a figure which shows an example of a code word. It is a flowchart which shows operation | movement of a detection apparatus. It is a flowchart which shows operation | movement of a detection apparatus. It is a flowchart which shows operation | movement of a detection apparatus. It is a figure for demonstrating operation | movement of a detection apparatus. It is a figure which shows the relationship between an image vector and a background vector. It is a figure which shows an example of a detection frequency map. It is a figure which shows an example of a non-detection frequency map. It is a flowchart which shows operation | movement of a detection apparatus. It is a figure which shows an example of an area | region where a detection target object has high possibility in an imaging area. It is a figure for demonstrating the adjustment method of a registration determination period. It is a figure which shows an example of the detection result of a detection target object. It is a figure which shows an example of a detection frequency map. It is a figure which shows an example of the detection result of a detection target object. It is a figure which shows an example of the detection result of a detection target object. It is a figure which shows an example of a detection frequency map. It is a figure which shows the example of an update of a detection frequency map. It is a figure which shows the example of an update of a non-detection frequency map. It is a figure which shows the example of an update of a detection frequency map.

<Outline of operation of detection device>
FIG. 1 is a diagram showing an outline of the operation of the detection apparatus 1 according to the embodiment. Based on the input image, the detection apparatus 1 detects a detection target existing in an imaging region that appears in the image, that is, in an imaging region (field-of-view range) of an imaging unit that captures the image.

In the present embodiment, the detection target is a moving object, more specifically a person. The detection apparatus 1 detects a moving object image (an image showing a moving object) included in an input image, thereby detecting a moving object, that is, a moving object existing in an imaging region that appears in the image.

As shown in FIG. 1, in the present embodiment, there are a preparation stage and an actual operation stage as the operation stage of the detection apparatus 1. In the preparation stage, the detection apparatus 1 executes a background model generation process using a plurality of input images sequentially input in time series. The background model is a model configured by collecting information included in a plurality of input images obtained by photographing the same scene (subject). The background model is used when a moving body image is detected from each input image sequentially input in time series in a moving body detection process described later. The preparation stage in which the background model generation process is executed is also referred to as a “learning stage”. Hereinafter, an input image used for generating a background model may be referred to as a “reference image”. In addition, an input image that is a detection target of a moving body image may be referred to as a “detection target image”.

In the detection apparatus 1, when the generation of the background model is completed, the operation stage shifts from the preparation stage to the actual operation stage. In the actual operation stage, the detection apparatus 1 detects a moving object image with respect to an input image, a background model update process for updating a background model, and a registration determination period described later used in the background model update process. And a determination period adjustment process for adjusting. In this embodiment, the detection target is a person, but it may be other than a person.

<Configuration of detection device>
FIG. 2 is a block diagram showing the configuration of the detection apparatus 1. As shown in FIG. 2, the detection device 1 includes an image input unit 2, an image processing unit 3, a detection result output unit 4, a background model storage unit 5, and a cache model storage unit 6. FIG. 3 is a block diagram showing a configuration of the image processing unit 3.

The image input unit 2 inputs an input image 200 input from the outside of the detection apparatus 1 to the image processing unit 3. The input image 200 is a captured image captured by the imaging unit. FIG. 4 is a diagram illustrating an example of the imaging region 10 that appears in the input image 200, that is, the imaging region (field-of-view range) 10 of the imaging unit that captures the input image 200. The imaging area 10 shown in FIG. 4 includes a conference room 100 as a subject. Therefore, in this case, the input image 200 input to the detection apparatus 1 is an image showing the conference room 100. In the conference room 100, a plurality of desks 101 and a plurality of chairs 102 are arranged so as to surround the center of the floor. The outside of the plurality of desks 101 is a passage 103. In the conference room 100, a curtain 104 is provided on a part of the wall. The detection apparatus 1 according to the present embodiment detects a person existing in the conference room 100 by detecting a moving body image with respect to the input image 200 showing the conference room 100, for example.

The image processing unit 3 performs various image processing on the input image 200 input from the image input unit 2. The image processing unit 3 includes a CPU 300 and a storage unit 310. The storage unit 310 includes a non-transitory recording medium that can be read by the CPU 300, such as a ROM (Read Only Memory) and a RAM (Random Access Memory). The storage unit 310 stores a control program 311 for controlling the detection device 1. Various function blocks are formed in the image processing unit 3 by the CPU 300 executing the control program 311 in the storage unit 310.

Note that the storage unit 310 may include a computer-readable non-transitory recording medium other than the ROM and RAM. The storage unit 310 may include, for example, a small hard disk drive and an SSD (Solid State Drive).

As shown in FIG. 3, the image processing unit 3 includes a plurality of functional blocks such as a background model generation unit 30, a moving object detection unit 31, a determination period adjustment unit 32, and a background model update unit 33. These functional blocks are not realized by the CPU executing the program, but may be realized by a hardware circuit using a logic circuit or the like that does not require a program for realizing the function.

The background model generation unit 30 generates a background model 500 including background image information using a plurality of input images 200 (a plurality of reference images 200) sequentially input from the image input unit 2. The background model 500 generated by the background model generation unit 30 is stored in the background model storage unit 5.

The moving object detection unit 31 detects a moving object image in the input image 200 using the input image 200 input from the image input unit 2 and the background model 500 in the background model storage unit 5. In other words, the moving object detection unit 31 uses the input image 200 and the background model 500 to detect a moving object that exists in the imaging region 10 that appears in the input image 200.

The background model update unit 33 updates the background model 500. The background model update unit 33 includes a specifying unit 330 and a registration determining unit 331. Based on the detection result of the moving object detection by the moving object detection unit 31, the specifying unit 330 specifies an area where the moving object is highly likely to exist in the imaging region 10. Specifically, the specifying unit 330 generates a detection frequency map and a non-detection frequency map based on the detection result of the moving object detection unit 31, and the moving object may exist in the imaging region 10 using these maps. Identify areas with high The detection frequency map shows a distribution of detection frequencies of moving objects (detection objects) in a plurality of areas in the imaging area 10. The non-detection frequency map indicates a distribution of non-detection frequencies of moving objects in a plurality of regions in the imaging region 10. The registration determination unit 331 determines whether or not to register the image information obtained from the input image 200 as background image information in the background model 500 using the specifying result in the specifying unit 330. The detection frequency map and the non-detection frequency map will be described in detail later.

The background model storage unit 5 stores the background model 500 generated by the background model generation unit 30. The cache model storage unit 6 stores a cache model to be described later. Each of the background model storage unit 5 and the cache model storage unit 6 includes rewritable storage means such as flash memory, EPROM (Erasable Programmable Read Only Memory), or hard disk (HD). In this example, the background model storage unit 5 and the cache model storage unit 6 are independent in hardware, but a part of the storage area of one storage device is used as the background model storage unit 5, Another part of the storage area may be used as the cache model storage unit 6.

The determination period adjustment unit 32 adjusts the registration determination period used in the update of the background model 500. The detection result output unit 4 outputs the detection result of the moving object detection by the moving object detection unit 31 to the outside. The detection result output unit 4 includes, for example, a display unit that displays the state of the imaging region 10, that is, the state of the subject existing in the imaging region 10 (the conference room 100 in the example of FIG. 4) in real time. The display unit displays the area where the moving object is detected in color or the like, so that the detection result of the moving object detection is output to the outside. In addition, the detection result output unit 4 may output the detection result to the outside with a sound such as voice. The detection result output unit 4 may output the detection result to the outside by outputting a signal indicating the detection result to an external device. In this case, the external device performs an operation according to the detection result. For example, the external device generates an alarm. Alternatively, when the imaging region 10 is the conference room 100 in FIG. 4, the external device controls the lighting apparatus in the conference room 100 to brighten only the region where a person exists. In addition, the external device controls the air conditioner in the conference room 100 to cool or warm only an area where a person exists.

<Preparation stage (background model generation process)>
Next, background model generation processing performed in the preparation stage of the detection apparatus 1 will be described. FIG. 5 is a diagram for explaining the background model 500. In the present embodiment, when there is no person in the imaging region 10 (when the conference room 100 is not used), the input image 200 captured by the imaging unit is used as the reference image 200 used for generating the background model 500. It becomes. In the background model generation process, a background model 500 is generated based on A (A ≧ 2) reference images 200.

In the present embodiment, the imaging area 10 is divided into a plurality of rectangular imaging blocks (partial imaging areas). When an area indicating an image of a certain imaging block included in the input image 200 is referred to as an “image block”, the input image 200 includes a plurality of image blocks respectively indicating images of a plurality of imaging blocks constituting the imaging area 10. Is done. In the present embodiment, the size of one image block is, for example, 3 pixels × 3 pixels. Hereinafter, regarding the imaging block and the image block indicating the image of the imaging block in the input image 200, the imaging block may be referred to as an imaging block corresponding to the image block.

As shown in FIG. 5, the background model 500 includes a plurality of codebooks CB corresponding to the plurality of imaging blocks BK, respectively. Each code book CB includes a code word CW including image information and related information related to the image information. The code word CW included in the code book CB is generated based on an image block indicating an image of the imaging block BK corresponding to the code book CB in one input image 200. Each code book CB includes a plurality of code words CW. Hereinafter, the image information included in the code word CW in the background model 500 may be referred to as “background image information”.

In FIG. 5, the code book CB showing sand hatching includes three code words CW1 to CW3 respectively generated based on the three reference images 200a to 200c. The code word CW1 included in the code book CB is generated based on the image block indicating the image of the imaging block BK corresponding to the code book CB in the reference image 200a. The code word CW2 included in the code book CB is generated based on the image block indicating the image of the imaging block BK corresponding to the code book CB in the reference image 200b. The code word CW3 included in the code book CB is generated based on the image block indicating the image of the imaging block BK corresponding to the code book CB in the reference image 200c.

FIG. 6 is a diagram for explaining the code word CW. In the code word CW, image information of an image block indicating an image of an imaging block corresponding to the code book CB including the code word CW, that is, pixel values PV of a plurality of pixels constituting the image block are used as background image information. include. The code word CW includes the latest matching time Te and the code word generation time Ti as related information. As will be described later, it is determined whether or not the image information in the code word CW included in the background model 500 matches the image information acquired from the detection target image 200. The latest matching time Te included in the code word CW indicates the latest time when it is determined that the image information included in the code word CW matches the image information acquired from the detection target image 200. The code word generation time Ti included in the code word CW indicates the time at which the code word CW is generated.

FIG. 7 is a flowchart showing background model generation processing in which such a background model 500 is generated. The background model generation process shown in FIG. 7 is executed when the background model 500 is not stored in the background model storage unit 5.

As illustrated in FIG. 7, when the reference image 200 is input from the image input unit 2 to the image processing unit 3 in step s1, the background model generation unit 30 selects an imaging block having the imaging region 10 in step s2. Let it be an attention imaging block. Then, the background model generation unit 30 determines whether or not the code book CB corresponding to the target imaging block is stored in the background model storage unit 5.

When the background model generation unit 30 determines that the code book CB corresponding to the target imaging block is not stored in the background model storage unit 5, the target model imaging is performed based on the reference image 200 input in step s1 in step s3. A code book CB corresponding to the block is generated and stored in the background model storage unit 5.

Specifically, the background model generation unit 30 acquires image information from an image block indicating an image of the target imaging block in the reference image 200 input in step s1. Then, the background model generation unit 30 generates a code word CW including the acquired image information as background image information, and stores the code book CB including the code word CW in the background model storage unit 5. The latest match time Te included in the code word CW is provisionally set to the same time as the code word generation time Ti.

On the other hand, when the background model generation unit 30 determines that the code book CB corresponding to the target imaging block is stored in the background model storage unit 5, the target model in the reference image 200 input in step s1 in step s4. Image information is acquired from an image block indicating an image of the imaging block. Then, the background model generation unit 30 matches the acquired image information with the background information image stored in the background model storage unit 5 and included in each code word CW included in the code book CB corresponding to the imaging block of interest. It is determined whether or not to do. That is, the background model generation unit 30 determines whether or not there is a code word CW including background image information that matches the acquired image information in the code word CW included in the code book CB corresponding to the imaging block of interest. .

If it is determined in step s4 that the background information image in each codeword CW included in the codebook CB corresponding to the target imaging block does not match the acquired image information in step s5, that is, the target If there is no code word CW including background image information that matches the acquired image information in the code word CW included in the code book CB corresponding to the imaging block, in step s6, the background model generation unit 30 performs step In step s4, a code word CW including the image information acquired from the reference image 200 as background image information is generated. The latest match time Te included in the code word CW is provisionally set to the same time as the code word generation time Ti. Then, the background model generation unit 30 adds the generated code word CW to the code book CB corresponding to the target imaging block stored in the background model storage unit 5. As a result, new background image information is added to the code book CB corresponding to the imaging block of interest. Thereafter, step s7 is executed.

On the other hand, when the background information image in the code word CW included in the code book CB corresponding to the imaging block of interest matches the acquired image information in step s5, that is, the code corresponding to the imaging block of interest. If there is a codeword CW including background image information that matches the acquired image information in the codeword CW included in the book CB, step s7 is executed without executing step s6.

In step s7, the background model generation unit 30 determines whether or not processing has been performed for all of the imaging blocks in the imaging region 10, that is, whether or not all of the imaging blocks have been set as the imaging block of interest. As a result of the determination in step s7, if there is an imaging block that has not been processed, the background model generation unit 30 sets the imaging block that has not been processed yet as a new attention imaging block, and thereafter step s2 and subsequent steps. Execute.

On the other hand, if the result of determination in step s7 is that processing has been performed for all of the imaging blocks in the imaging area 10, the background model generation unit 30 applies to the A reference images 200 in step s8. It is determined whether or not similar processing has been performed. If the result of determination in step s8 is that the number of processed reference images 200 is smaller than A, the background model generation unit 30 newly inputs a reference to the image processing unit 3 in step s1. The process from step s2 is executed on the image 200. If the result of determination in step s8 is that the number of processed reference images 200 is A, the background model generation unit 30 ends the background model generation process. Thereby, the background model 500 as described above is generated in the background model storage unit 5.

<Real operation stage>
Next, the operation of the detection device 1 in the actual operation stage will be described. FIG. 8 is a flowchart showing a schematic operation in the actual operation stage of the detection apparatus 1. In the detection apparatus 1, when the background model generation process ends, the process shown in FIG. 8 is executed.

As shown in FIG. 8, when the input image 200 is input from the image input unit 2 to the image processing unit 3 in step s11, a series of processing from steps s12 to s14 is executed with the input image 200 as a processing target. Is done.

In step s12, the image processing unit 3 performs a moving object detection process for detecting a moving object image on the input image 200 to be processed. In step s13, the image processing unit 3 performs a determination period adjustment process for adjusting the registration determination period based on the result of the moving object detection process in step s12. Thereafter, in step s14, the image processing unit 3 performs background model update processing for updating the background model 500 in the background model storage unit 5.

Thereafter, when a new input image 200 (new detection target image 200) is input from the image input unit 2 to the image processing unit 3 in step s11, the input image 200 is set as a new processing target, and steps s12 to s14 are performed. A series of processes up to are executed. Thereafter, the image processing unit 3 operates in the same manner.

As described above, in the detection apparatus 1 according to the present embodiment, each time the input image 200 is input, the moving object detection process, the determination period adjustment process, and the background model update process are executed in this order.

<Moving object detection processing>
Next, the moving object detection process in step s12 will be described in detail. FIG. 9 is a flowchart showing the moving object detection process. As illustrated in FIG. 9, in step s121, the moving object detection unit 31 sets an imaging block having the imaging region 10 (for example, an upper left imaging block in the imaging region 10) as a target imaging block. Then, the moving object detection unit 31 may be referred to as an image block (hereinafter referred to as “attention image block”) indicating an image of the target imaging block in the processing target input image 200 (detection target image 200) input in step s11. ) Is detected. That is, the moving object detection unit 31 detects whether a moving object exists in the imaging block of interest.

In the moving object detection according to the present embodiment, the image information acquired from the target image block in the input image 200 and the background in each code word CW included in the code book CB corresponding to the target imaging block in the background model 500. By determining whether or not the image information matches, it is determined whether or not the target image block is a moving image. Hereinafter, the code book CB corresponding to the imaging block of interest may be referred to as a “corresponding code book CB”. Further, the code word CW included in the corresponding code book CB may be referred to as “corresponding code word CW”. A specific method of moving object detection will be described later.

When step s121 is executed, in step s122, the moving object detection unit 31 stores the result of the moving object detection in step s121. Then, in step s123, the moving object detection unit 31 determines whether or not processing has been performed for all imaging blocks in the imaging region 10, that is, whether or not all imaging blocks have been set as the imaging block of interest. As a result of the determination in step s123, when there is an imaging block that has not been processed, the moving object detection unit 31 sets the imaging block that has not been processed yet as a new attention imaging block, and thereafter performs step s121 and subsequent steps. Execute. On the other hand, if the result of determination in step s123 is that processing has been performed for all imaging blocks in the imaging area 10, that is, detection of moving object images has been completed for all areas of the input image 200. In this case, the moving object detection unit 31 ends the moving object detection process. As a result, the moving object detection unit 31 stores the results of moving object detection for a plurality of image blocks constituting the input image 200. That is, the moving object detection unit 31 stores the results of moving object detection for a plurality of imaging blocks constituting the imaging area 10. This detection result is input to the detection result output unit 4.

<Details of motion detection>
Next, a specific method of moving object detection in step s121 will be described with reference to FIGS. FIG. 10 is a diagram showing how vectors are extracted from each of the target image block of the input image 200 and the corresponding codeword CW of the background model 500. FIG. 11 is a diagram illustrating a relationship between a vector extracted from the target image block of the input image 200 and a vector extracted from the corresponding codeword CW of the background model 500.

In this embodiment, the image information of the target image block in the input image 200 is handled as a vector. For each corresponding codeword CW in the background model 500, the background image information included in the corresponding codeword CW is treated as a vector. Based on whether or not the vector for the image information of the target image block and the vector for the background image information of each corresponding codeword CW are in the same direction, whether or not the target image block is a moving image. Is determined. When these two types of vectors are directed in the same direction, it can be considered that the image information of the target image block and the background image information of each corresponding codeword CW match. Therefore, in this case, the target image block in the input image 200 is determined not to be a moving body image as it is the same as the image indicating the background. On the other hand, when the two types of vectors do not point in the same direction, it can be considered that the image information of the target image block does not match the background image information of each corresponding codeword CW. Therefore, in this case, it is determined that the target image block in the input image 200 is not an image showing the background but a moving body image.

Specifically, the moving object detection unit 31 generates an image vector x _f in which the pixel values of a plurality of pixels included in the image block of interest in the input image 200 and components. Figure 10 shows a image vector x _f which was a pixel value of each pixel component of the target image block 210 having nine pixels. In the example of FIG. 10, each pixel has pixel values of R (red), G (green), and B (blue), so the image vector _xf is composed of 27 components.

Similarly, the moving object detection unit 31 uses the background image information in the corresponding codeword CW included in the corresponding codebook CB of the background model 500 to generate a background vector that is a vector related to the background image information. The background image information 510 of the corresponding codeword shown in FIG. 10 includes pixel values for nine pixels. Therefore, a background vector _xb having the pixel values for the nine pixels as components is generated. The background vector _xb is generated from each of a plurality of code words CW included in the corresponding code book CB. Therefore, a plurality of background vector x _b are generated for one image vector x _f.

As described above, when the image vector _xf and each background vector _xb are in the same direction, the target image block in the input image 200 is not different from the image indicating the background. However, since the image vector _xf and each background vector _xb are considered to contain a certain amount of noise components, the image vector _xf and each background vector _xb are not completely in the same direction. However, it is possible to determine that the target image block in the input image 200 is an image indicating the background.

Therefore, in the present embodiment, the image vector _xf and each background vector _xb are completely the same in consideration that the image vector _xf and each background vector _xb include a certain amount of noise components. Even if it is not facing the direction, it is determined that the target image block in the input image 200 is an image indicating the background.

Assuming that contains image vector x _f and background vector x _b in the noise component, the relationship between the image vector x _f and background vector x _b to the true vector u can be expressed as in FIG. 11. In this embodiment, the image and the vector x _f and background vector x _b is, as an evaluation value that indicates whether the pointing how the same direction, consider the evaluation value D ² represented by the following (1).

Then, the matrix X by using the image vector x _f and background vector x _b, expressed by the equation (2), evaluation value D ² is a minimum eigenvalue of a non-zero 2 × 2 matrix XX ^T. Accordingly, the evaluation value D ² can be determined analytically. Note that the evaluation value D ² is the minimum eigenvalue of the non-zero 2 × 2 matrix XX ^T is described in Non-Patent Document 3 above.

As described above, since the plurality of background vector x _b are generated for one image vector x _f, the value of evaluation values D ² which is represented using an image vector x _f and the background vector x _b As many as the number of background vectors _xb are obtained.

Decision image block of interest is whether the moving object image in the input image 200, the minimum value C of the plurality of values of the evaluation value D ^2, the mean value for a plurality of values of the evaluation value D ² mu and The moving object judgment formula shown by the following formula (3) expressed using the standard deviation σ is used. This moving object judgment formula is called Chebyshev's inequality.

Here, k in Expression (3) is a constant, and is a value determined based on an imaging environment (an environment in which the imaging unit is installed) of the imaging unit that captures the input image 200 and the like. The constant k is determined by experiments or the like.

When the moving object detection unit 31 satisfies the moving object determination formula (inequality), the moving image detection unit 31 considers that the image vector x _f and each background vector x _b do not face the same direction, and the target image block is not an image indicating the background, It determines with it being a moving body image. On the other hand, the moving object detection unit 31 considers that the image vector _xf and each background vector _xb face the same direction when the moving object determination formula is not satisfied, and the target image block is not a moving object image but a background. It is determined that the image is shown.

Thus, in the present embodiment, the moving object detection is performed based on whether the direction of the image vector obtained from the target image block and the direction of the background vector obtained from each corresponding codeword CW are the same. Therefore, the moving object detection method according to the present embodiment is a moving object detection method that is relatively robust against changes in brightness in the imaging region 10 such as sunshine changes or illumination changes.

<Detection frequency map and non-detection frequency map>
12 and 13 are diagrams showing examples of the detection frequency map 600 and the non-detection frequency map 610, respectively. The detection frequency map 600 includes the detection frequency 601 of the moving object (detection target) 700 for each of the plurality of imaging blocks BK constituting the imaging region 10. On the other hand, the non-detection frequency map 610 includes a non-detection frequency 611 of the moving object (detection target) 700 for each of the plurality of imaging blocks BK constituting the imaging region 10.

In the detection frequency map 600, a plurality of detection frequencies 601 included therein are arranged in a matrix. The detection frequency 601 of the moving object 700 for a certain imaging block BK is arranged at the same position as the position of the imaging block BK in the imaging area 10 in the detection frequency map 600. Similarly, in the non-detection frequency map 610, a plurality of non-detection frequencies 611 included therein are arranged in a matrix. The non-detection frequency 611 of the moving object 700 for a certain imaging block BK is arranged at the same position as the position of the imaging block BK in the imaging area 10 in the non-detection frequency map 610.

If the moving part detection unit 31 determines that an image block included in the input image 200 is a moving body image, that is, the specifying unit 330 determines that the moving body 700 exists in the imaging block corresponding to the image block. In the detection frequency map 600, the detection frequency 601 of the moving object 700 for the imaging block BK corresponding to the image block is increased by one.

If the moving object detection unit 31 determines that an image block included in the input image 200 is not a moving object image, that is, the specifying unit 330 determines that the moving object 700 does not exist in the imaging block corresponding to the image block. Then, the non-detection frequency 611 of the moving object 700 for the imaging block BK corresponding to the image block in the non-detection frequency map 610 is increased by one.

12 and 13, a detection frequency map generated when the moving object (person) 700 is stationary in two imaging blocks BK on the lower side of the center among the plurality of imaging blocks BK constituting the imaging region 10. An example of 600 and a non-detection frequency map 610 are shown. In the detection frequency map 600 shown in FIG. 12, the detection frequency 601 of the moving object 700 for the two imaging blocks BK where the moving object 700 exists is large. On the other hand, in the non-detection frequency map 610 shown in FIG. 13, the non-detection frequency 611 of the moving object 700 for each imaging block BK other than the two imaging blocks BK in which the moving object 700 exists is large.

In such a detection frequency map 600, the detection frequency 601 for the imaging block BK in which the moving body 700 is frequently detected increases. Therefore, by referring to the detection frequency map 600, it is possible to identify an area where the moving object 700 is likely to exist in the imaging area 10. In the imaging region 10, for example, when a person's passage exists, the detection frequency of the imaging block corresponding to the passage increases, and it can be understood that there is a high possibility that a person exists in the imaging block.

Further, in the non-detection frequency map 610, the non-detection frequency 611 for the imaging block BK in which the moving body 700 is not detected so much increases. Therefore, by referring to the non-detection frequency map 610, it is possible to specify an area in which there is a high possibility that the moving object 700 does not exist in the imaging area 10.

<Background model update process>
Next, the background model update process in step s14 will be described. In the background model update process, a cache model storage unit 6 that stores a cache model is used. The cache model includes background image information candidates that are background image information candidates registered in the background model 500.

Here, in the imaging area 10, the brightness may change due to a change in sunlight or a change in illumination. When the brightness in the imaging region 10 changes, the image information of the input image 200 changes, so the moving object detection unit 31 erroneously determines that the image block indicating the background included in the input image 200 is a moving object image. there's a possibility that. Therefore, there is a possibility that the image information of the image block determined to be a moving image by the moving object detection unit 31 is actually background image information.

Therefore, in the present embodiment, the background model update unit 33 once registers the image information of the image block determined to be a moving image by the moving object detection unit 31 as a background image information candidate in the cache model. Then, the background model update unit 33 determines whether the background candidate image information registered in the cache model is background image information based on the plurality of input images 200 input during the registration determination period. When the background model update unit 33 determines that the background image information candidate registered in the cache model is background image information, the background model update unit 33 registers the background image information candidate in the background model 500 as background image information. That is, the background model update unit 33 registers the background image information candidates stored in the cache model storage unit 6 in the background model 500 as background image information based on a plurality of input images 200 input during the registration determination period. Determine whether or not. The registration determination period is adjusted by the determination period adjustment process in step s13.

FIG. 14 is a flowchart showing the background model update process. As illustrated in FIG. 14, in step s141, the background model update unit 33 sets an imaging block having the imaging region 10 as a target imaging block. Then, the background model update unit 33 determines in the moving object detection unit 31 that the image block (attention image block) indicating the image of the target imaging block in the processing target input image 200 input in step s11 is a moving object image. It is determined whether or not. If it is determined in step s141 that the target image block is determined not to be a moving image by the moving object detection unit 31, that is, the image information of the target image block is the background image information of each corresponding codeword CW in the background model 500. If determined to match, the background model update unit 33 executes step s142.

In step s142, the background model update unit 33 changes the latest match time Te of the code word CW in the background model 500 including the background image information determined to match the image information of the target image block to the current time. When step s142 is executed, in the background model update unit 33, the specifying unit 330 increases the non-detection frequency for the target imaging block corresponding to the target image block in the non-detection frequency map 610 by one in step s143. .

On the other hand, when it is determined in step s141 that the target image block is determined to be a moving image by the moving object detection unit 31, the background model update unit 33 executes step s144. In step s144, the cache model is updated. Specifically, when the image information of the target image block is not included in each corresponding codeword CW included in the cache model in the cache model storage unit 6, the background model update unit 33 displays the image information. A code word CW included as a background image information candidate is generated and registered in the corresponding code block CB in the cache model. The code word CW includes the latest match time Te and the code word generation time Ti in addition to the image information (background image information candidate). The latest matching time Te included in the code word CW generated in step s144 is provisionally set to the same time as the code word generation time Ti. Further, the background model update unit 33, when the image information of the target image block is included in the corresponding codeword CW included in the cache model in the cache model storage unit 6, that is, the image information of the target image block When the corresponding code word CW including the matching background image information candidate is included in the cache model, the latest matching time Te in the code word CW including the background image information candidate in the cache model is changed to the current time. .

Thus, in step s144, the code word CW including the missing image information is added to the cache model, or the latest match time Te of the code word CW in the cache model is updated.

In step s144, when the code book CB corresponding to the target imaging block is not registered in the cache model in the cache model storage unit 6, the background model update unit 33 uses the image information of the target image block as the background. A code word CW included as an image information candidate is generated, and a code book CB including the code word CW is generated and registered in the cache model.

When step s144 is executed, the specifying unit 330 increases the detection frequency for the target imaging block corresponding to the target image block in the detection frequency map 600 by one in step s145.

When step s143 or step s145 is executed, in step s146, the background model update unit 33 determines whether or not processing has been performed for all of the imaging blocks in the imaging region 10, that is, sets all of the imaging blocks as the target imaging block. It is determined whether or not it has been set. If it is determined in step s146 that there is an imaging block that has not been processed, the background model update unit 33 sets the imaging block that has not been processed yet as a new target imaging block, and thereafter performs steps s141 and thereafter. Execute. On the other hand, if it is determined in step s146 that the processing has been performed for all the imaging blocks in the imaging region 10, the background model update unit 33 executes step s147.

In step s147, the code word CW that is included in the cache model and for which the latest match time Te has not been updated for a predetermined period is deleted. That is, when the image information included in the code word CW in the cache model does not match the image information acquired from the input image 200 for a certain period, the code word CW is deleted. If the image information included in the code word CW is background image information, that is, if it is image information obtained from an image indicating the background included in the input image 200, the code word CW includes The latest match time Te is frequently updated. Therefore, it can be considered that the image information included in the code word CW whose latest match time Te has not been updated for a predetermined period is highly likely to be image information acquired from the moving object image included in the input image 200. By deleting the code word CW whose latest match time Te has not been updated for a predetermined period from the cache model, the image information of the moving image is deleted from the cache model. Hereinafter, this predetermined period may be referred to as “deletion determination period”. The deletion determination period includes a change in brightness in the imaging region 10 such as a change in sunlight or a change in illumination, a change in image information due to a change in environment such as a poster installation or a desk layout change, and a person to be detected. This is a period set in advance to distinguish the change in image information that occurs when a moving object such as the moving object moves. For example, if the imaging frame rate of the imaging unit that captures the input image 200 is 30 fps and the imaging region 10 is the conference room 100 (see FIG. 4), the deletion determination period is from several tens of frames to several hundreds of frames. Is set to a period during which the input image 200 is input.

In step s147, when the code word CW that is included in the cache model and whose latest match time Te is not updated for the deletion determination period is deleted, the background model update unit 33 executes step s148. In step s148, the background model update unit 33 identifies the codeword CW that has been registered in the cache model and has passed the registration determination period from the codewords CW registered in the cache model. In step s144, when the code word CW is generated, the code word CW is immediately registered in the cache memory. Therefore, the code word CW is included in the code word CW as the time when the code word CW is registered in the cache model. The code word generation time Ti can be used.

The registration judgment period is set to a larger value than the deletion judgment period. The registration determination period is set to a value several times larger than the deletion determination period, for example. In the present embodiment, the registration determination period is represented by the number of frames. If the registration determination period is, for example, “500”, the registration determination period is a period in which the input image 200 for 500 frames is input.

When step s148 is executed, in step s149, the background model update unit 33 performs background model registration determination processing using the detection frequency map 600. In the background model registration determination process, it is determined whether or not to register the code word CW specified in step s148 in the background model 500 in the background model storage unit 5. The background model registration determination process will be described in detail below.

In the background model registration determination process, first, the specifying unit 330 uses the detection frequency map 600 to specify an area in the imaging area 10 where a moving object is highly likely to exist. Specifically, the specifying unit 330 specifies a detection frequency greater than the first threshold in the detection frequency map 600. Then, the specifying unit 330 sets an imaging block corresponding to a detection frequency larger than the first threshold as an area where there is a high possibility that a moving object exists in the imaging area 10. In other words, when the moving object detection frequency for the imaging block is greater than the first threshold, the specifying unit 330 determines that there is a high possibility that a moving object exists in the imaging block.

FIG. 15 is a diagram showing an example of the detection frequency map 600. When the first threshold value = 100, in the detection frequency map 600 shown in FIG. 15, the two detection frequencies in the first and second rows from the bottom of the rightmost column are higher than the first threshold value. It is getting bigger. Therefore, in the imaging region 10, there is a high possibility that a moving object is present in the imaging region 10 (an area indicated by diagonal lines) including the imaging blocks BK in the first and second rows from the bottom of the rightmost column. Region 120 becomes.

Note that when the detection frequency map 600 does not have a detection frequency greater than the first threshold, the specifying unit 330 determines that there is no region in the imaging region 10 where there is a high possibility that a moving object exists. . Further, the specifying unit 330 may set an imaging block corresponding to a detection frequency equal to or higher than the first threshold as an area where there is a high possibility that a moving object exists in the imaging area 10.

When the identification unit 330 identifies an area in the imaging area 10 where a moving object is highly likely to exist, the registration determination unit 331 uses the codeword CW for the background model 500 for each codeword CW identified in step s148. It is determined based on the identification result in the identification unit 330 whether or not to be registered.

Specifically, the registration determining unit 331 determines that the imaging block corresponding to the image block from which the image information included in the code word CW identified in step s148 is acquired is identified in the imaging region 10 identified by the identifying unit 330. If it is included in the region where the moving object is highly likely to exist, that is, if the moving object is highly likely to exist in the imaging block, it is determined not to register the code word CW in the background model 500.

On the other hand, the registration determination unit 331 determines that an imaging block corresponding to the image block from which the image information included in the code word CW identified in step s148 is acquired is identified by the identification unit 330. If it is not included in the region that is highly likely to exist, that is, if the possibility that a moving object is not present in the imaging block is high, it is determined that the code word CW is registered in the background model 500.

If the specifying unit 330 determines that there is no region where there is a high possibility that a moving object exists in the imaging region 10, the registration determining unit 331 uses all the code words CW specified in step s148 as the background. It is determined that the model 500 is registered.

The background model update unit 33 registers the codeword CW determined to be registered in the background model 500 by the registration determination unit 331 in the code block CB corresponding to the codeword CW in the background model 500. Then, the background model update unit 33 deletes the codeword CW registered in the background model 500 from the cache model.

As can be understood from the above description, in the present embodiment, the background model update unit 33 may delete the code word CW in the cache model until the registration determination period elapses after it is registered in the cache model. . The background model update unit 33 deletes the code word CW (background image information candidate) in the cache model until the registration determination period elapses after it is registered in the cache model. This means that it is determined that the code word CW (background image information candidate) registered in the cache memory is not registered in the background model 500 based on a plurality of input images 200 input in the determination period.

In the present embodiment, the background model updating unit 33 registers the code word CW in the cache model in the background model 500 without deleting it until the registration determination period elapses after the registration in the cache model. There is. The background model update unit 33 registers the codeword CW (background image information candidate) in the cache model in the background model 500 without deleting it until the registration determination period elapses after it is registered in the cache model. This means that the model updating unit 33 has determined that the code word CW (background image information candidate) registered in the cache memory is registered in the background model 500 based on a plurality of input images 200 input in the registration determination period. ing.

Thus, the background model update unit 33 registers the background image information candidate registered in the cache model as the background image information in the background model 500 based on the plurality of input images 200 input in the registration determination period. It is determined whether or not. Therefore, image information of an image block erroneously determined to be a moving image by the moving object detection unit 31 can be registered in the background model 500 as background image information. Therefore, the background model 500 can be updated appropriately, and the accuracy of moving object detection by the moving object detection unit 31 is improved.

Further, unlike the present embodiment, among the codewords CW registered in the cache model, the codeword CW that has passed the registration determination period after being registered in the cache model is unconditionally registered in the background model 500. In this case, there is a possibility that the code word CW registered in the cache model and including the image information of the moving object image is registered in the background model 500. Specifically, when the moving object is temporarily stationary in the imaging region 10, the image information of the image block indicating the image of the imaging block in which the moving object exists in the imaging region 10 is unlikely to change. Accordingly, there is a possibility that the code word CW registered in the cache model and including the image information of the moving image indicating the moving object is not deleted from the cache model for a long time and is finally registered in the background model 500. . Thereby, the precision of the moving body detection in the moving body detection part 31 may deteriorate. In particular, as will be described later, when the registration determination period is adjusted to be shortened, the code word CW including the image information of the moving body image registered in the cache model is easily registered in the background model 500. As a result, the possibility that the accuracy of moving object detection by the moving object detection unit 31 deteriorates increases.

Therefore, in the present embodiment, the detection frequency map 600 is used as described above, instead of unconditionally registering the codeword CW, which has been registered in the cache model and having passed the registration determination period, in the background model 500. Whether the code word CW is registered in the background model 500 is determined. Thereby, it can suppress that the codeword CW containing the image information of a moving body image is registered into the background model 500 accidentally. This point will be described in detail below.

As described above, the specifying unit 330 uses the detection frequency map 600 to specify an imaging block that is highly likely to have a moving object in the imaging region 10. There is a high possibility that the image information of the image block indicating the image of the imaging block that is highly likely to be present in the input image 200 is the image information of the moving object image. Therefore, even if the code word CW has been registered for the cache model and the registration determination period has elapsed, the imaging block corresponding to the image block from which the image information included in the code word CW is acquired has moving objects. If the imaging block has a high possibility, the image information of the code word CW is highly likely to be image information of a moving body image. Therefore, as in the present embodiment, even a code word CW for which the registration determination period has elapsed since registration in the cache model corresponds to the image block from which the image information included in the code word CW has been acquired. If the imaging block is an imaging block in which there is a high possibility that a moving object exists, the code word CW including the image information of the moving object image is erroneously added to the background model 500 by not adding the code word CW to the background model 500. It can be suppressed.

When step s149 ends, in step s150, the background model update unit 33 performs detection frequency clear processing. In the detection frequency clear process, first, the specifying unit 330 specifies a non-detection frequency larger than the second threshold in the non-detection frequency map 610. Then, the specifying unit 330 clears and sets the detection frequency in the detection frequency map 600 for the imaging block corresponding to the non-detection frequency larger than the second threshold to zero. Furthermore, the specifying unit 330 clears a non-detection frequency larger than the second threshold in the non-detection frequency map 610 and sets it to zero. For example, the second threshold value is set to the same value as the first threshold value.

Here, the detection frequency increases each time the presence of a moving object in the imaging block corresponding to the detection frequency is detected in the moving object detection process in step s12. Therefore, when the detection frequency is not cleared, there is a state in which the detection frequency of an imaging block for which there is no high possibility that a moving object currently exists due to a layout change or the like in the imaging region 10 is higher than the first threshold value. It is more likely to occur.

On the other hand, when a certain imaging block transitions from a state in which there is a high possibility that a moving object exists to a state where it does not, the non-detection frequency of the imaging block in the non-detection frequency map 610 increases.

Therefore, in the present embodiment, the detection frequency in the detection frequency map 600 for the imaging block corresponding to the non-detection frequency larger than the second threshold in the non-detection frequency map 610 is cleared. As a result, it is possible to suppress the occurrence of a state in which the detection frequency of an imaging block for which there is no high possibility that a moving object currently exists is greater than the first threshold value. Therefore, it is possible to accurately specify an area in the imaging area 10 where a moving object is highly likely to exist. Therefore, it can further suppress that the image information of a moving body image is registered into the background model 500. FIG. As a result, the accuracy of moving object detection is further improved. The specifying unit 330 may clear the detection frequency and the non-detection frequency for the imaging block corresponding to the non-detection frequency equal to or higher than the second threshold.

When step s150 ends, in step s151, the background model update unit 33 includes a codeword CW including background image information that does not match the image information of the input image 200 for a predetermined period in the codeword CW included in the background model 500. Is deleted. That is, the background model update unit 33 deletes the code word CW included in the background model 500 and whose latest match time Te has not been updated for a predetermined period. Thereby, in the imaging region 10, the code word CW including the image information acquired from the image of the imaging block that is no longer the background due to the time-series imaging environment change can be deleted from the background model 500. Therefore, the information amount of the background model 500 can be reduced.

Even if a change in the imaging environment such as a change in brightness in the imaging region 10 occurs by performing such a background model update process, the background model 500 that follows the change in the imaging environment is used. Moving object detection can be performed. Therefore, the accuracy of moving object detection is improved.

In the present embodiment, the detection frequency map 600 and the non-detection frequency map 610 are updated in the background model update process. Instead, after the moving object detection process in step s12 ends, the step Before the background model update process in s14 starts, the detection frequency map 600 and the non-detection frequency map 610 may be updated based on the result of moving object detection for each imaging block in the moving object detection unit 31. For example, the detection frequency map 600 and the non-detection frequency map 610 may be updated between the moving object detection process in step s12 and the determination period adjustment process in step s13.

<Determination period adjustment process>
When the brightness in the imaging region 10 changes suddenly due to changes in sunlight or illumination, the image information of the input image 200 changes suddenly. Therefore, it may be erroneously determined that the image indicating the background included in the input image 200 is a moving image, and the image information of the image indicating the background may be registered in the cache model. In such a case, if the registration determination period used in the background model update process is long, the background image information in the cache model is not reflected in the background model 500 for a long time. As a result, the accuracy of moving object detection may deteriorate.

Therefore, the registration determination period is adjusted in the determination period adjustment process in step s13. Specifically, the determination period adjustment unit 32 sets the registration determination period to the moving object region (the region determined to be a moving object image by the moving object detection unit 31) in the processing target input image 200 input in step s11. The smaller the ratio, the smaller. Thereby, when the ratio of the moving body area | region in the input image 200 is large, the registration determination period used by the background model update process in step s14 becomes short. When the brightness in the imaging region 10 changes suddenly, the image information changes suddenly in the entire input image 200, so the proportion of the moving object region in the input image 200 increases. Therefore, when the brightness in the imaging region 10 changes suddenly, the registration determination period used in the background model update process is shortened. As a result, the background image information in the cache model can be immediately reflected in the background model, and the accuracy of moving object detection is improved.

In the present embodiment, the registration determination period Dt is expressed by the following formula (4), where Rd is the ratio of the moving object region in the input image 200.

Here, a in equation (4) is a threshold value, and Dmin and Dmax are constants. Dmax> Dmin.

FIG. 16 shows the relationship between the registration determination period Dt and the ratio Rd of the moving object area in the input image 200 expressed by the equation (4). As shown in FIG. 16, according to the equation (4), the registration determination period Dt becomes shorter as the moving object region ratio Rd in the input image 200 becomes larger. In particular, when the ratio Rd of the moving object region in the input image 200 exceeds a fixed threshold value a, the registration determination period Dt becomes the minimum value (Dmin).

It should be noted that the threshold value a is considered abnormal when it is determined that the region of the input image 200 or more of the input image 200 is a moving body image (whether it is considered that the brightness in the imaging region 10 has suddenly changed). ), Which is set in advance based on the criterion of), and is set according to the subject in the imaging region 10.

The ratio Rd of the moving object region in the input image 200 is expressed by the following equation (5), where Pd is the number of pixels in the moving object region and Pa is the total number of pixels in the input image 200.

The number of pixels Pd in the moving object region can be obtained by multiplying the number of image blocks that are moving object images by the number of pixels included in one image block.

<Relationship between Registration Determination Period and First Threshold Regarding Detection Frequency Map>
When the number of frames representing the registration determination period is small, before the detection frequency of the imaging block is higher than the first threshold for an imaging block in which there is a high possibility of moving objects in the imaging region 10, There is a high possibility that a code word CW including image information acquired from an image block indicating an image of the imaging block in the cache model is registered in the background model 500. Therefore, when the number of frames representing the registration determination period is small, the code word CW including the image information of the moving body image in the cache model may be erroneously added to the background model 500. In order to suppress this, it is conceivable that Dmin indicating the minimum value of the registration determination period is larger than the first threshold value.

However, if Dmin is increased, the above-mentioned effect that “the image information of the background in the cache model can be immediately reflected in the background model” is diminished.

Therefore, in the present embodiment, the first threshold value is set to Dmin <first threshold value <Dmax. For example, the first threshold value is half of the value obtained by adding Dmin and Dmax. Thereby, when the ratio Rd of the moving body region in the input image 200 is small, the registration determination period becomes longer than the first threshold value. Therefore, when the ratio Rd of the moving object region in the input image 200 is small, it is possible to prevent the code word CW including the image information of the moving object image in the cache model from being erroneously added to the background model 500. it can. On the other hand, when the ratio Rd of the moving object region in the input image 200 is large, the registration determination period is smaller than the first threshold value. Therefore, the background image information in the cache model can be registered immediately.

As described above, in the present embodiment, the identification result of the region where the detection target object in the imaging region 10 is likely to exist is used, and the image information obtained from the input image 200 is used as the background image information as the background model. Whether or not to register in 500 is determined. Therefore, it is possible to suppress registration of image information obtained from an image of an area where the detection target actually exists in the imaging area 10 as background image information in the background model 500. Therefore, the detection accuracy of the detection object is improved.

FIG. 17 is a diagram illustrating an example of a result of moving object detection by the moving object detection unit 31. In FIG. 17, a moving object region 800 detected from the input image 200 by the moving object detection unit 31 is shown superimposed on the input image 200. The input image 200 shown in FIG. 17 includes an image of a room 900 in which a plurality of cardboards 920 and a plurality of chairs 930 are arranged on a floor 910 as an image of a subject. A window (not shown) is provided on the wall 940 of the room 900. Two

people

990a and 990b exist in the room 900. In the example of FIG. 17, each of the

people

990a and 990b existing in the room 900 is appropriately detected as the moving object region 800.

FIG. 18 is a diagram showing a detection frequency map 600 when the

persons

990a and 990b are stationary at the position shown in FIG. In FIG. 18, the floor 910 and the wall 940 are indicated by broken lines. In FIG. 18, for easy understanding, the detection frequency map 600 is shown by dividing the magnitude of the detection frequency into, for example, three stages from the first stage to the third stage. In the detection frequency map 600 shown in FIG. 18, the region belonging to the third stage where the detection frequency is the highest shows hatching leftward, and the region belonging to the second largest second stage is raised rightward. The hatching is shown. In the detection frequency map 600 shown in FIG. 18, hatching is not shown for the region belonging to the first stage with the lowest detection frequency. In the detection frequency map 600 shown in FIG. 18, the detection frequency for the region where the

people

990a and 990b exist in the imaging region 10 is large.

FIG. 19 is a diagram illustrating an example of a result of moving object detection in the moving object detection unit 31 when the detection frequency map 600 is not used for updating the background model 500, unlike the present embodiment. In FIG. 19, a moving object region 800 detected by the moving object detection unit 31 from the input image 200 is shown superimposed on the input image 200. In the input image 200 shown in FIG. 19, two

people

990a and 990b are sitting on two chairs 930, respectively. In the moving object detection result shown in FIG. 19, the person 990a is detected as the moving object area 800, but the person 990b is not detected as the moving object area 800.

FIG. 20 is a diagram illustrating an example of a moving object detection result in the present embodiment, that is, a result of moving object detection in the moving object detection unit 31 when the detection frequency map 600 is used to update the background model 500. . As shown in FIG. 20, when the detection frequency map 600 is used to update the background model 500, each of the

people

990 a and 990 b is appropriately detected as the moving object region 800.

When the environment of the imaging region 10 is an environment where there is little movement of moving objects such as a conference room, the first threshold value for the detection frequency map 600 is the second threshold value for the non-detection frequency map 610. It is desirable to be larger than the threshold value. In an environment where there is little movement of a moving object, there is a high possibility that the moving object stays in the same place, and therefore the detection frequency of an imaging block that is highly likely to exist is likely to increase. In such an environment, if the first threshold value is decreased, there is a possibility that it is erroneously determined that there is a high possibility that a moving object exists for an imaging block in which the moving object happens to stay a little. On the other hand, if the second threshold value is increased, even if the imaging block changes to an area where the possibility that a moving object exists is not high due to a change in the environment, the detection frequency of the imaging block is forever. May not be cleared. Therefore, when the environment of the imaging region 10 is an environment in which the movement of the moving object is small, it is desirable that the first threshold value is large and the second threshold value is small.

In addition, when the environment of the imaging region 10 is an environment where there is a lot of movement of moving objects such as a place where there is a lot of traffic, the first threshold value is preferably smaller than the second threshold value. In an environment where there is a lot of movement of moving objects, it is unlikely that the moving objects will stay in the same place. Therefore, even in an imaging block where there is a high possibility that a moving object exists, the detection frequency of the imaging block is unlikely to increase. In such an environment, if the first threshold value is increased, there is a possibility that a region where a moving object is likely to exist cannot be appropriately specified. On the other hand, if the second threshold value is decreased, the moving object may move slightly from the imaging block, and the detection frequency for the imaging block may be cleared. Therefore, when the environment of the imaging region 10 is an environment where there is a lot of movement of moving objects, it is desirable that the first threshold value is small and the second threshold value is large.

<Modification>
Even if a moving body such as a person is stationary, the moving body is rarely completely stationary and may move slightly. For example, a person sitting at a desk or talking to another person while standing in a passage does not change its overall position (although it is macroscopically stationary), but has a hand, head The body, legs, etc. move slightly.

FIG. 21 is a diagram illustrating an example of a state in which the person 990 slightly moves in the imaging region 10 and an example of a detection frequency map 600 in that case. As shown in FIG. 21, when the person 990 moves slightly, the detection frequency map 600 increases the detection frequency for the imaging block in which the person 990 mainly exists, but the surrounding imaging blocks of the imaging block. The detection frequency for is not so great. Accordingly, since the image information of the moving body image acquired from the image block corresponding to the imaging block around the imaging block in the cache model has a low detection frequency of the imaging block around the imaging block, the background model 500 may be registered.

Therefore, in the present modification, as illustrated in FIG. 22, in step s145, the specifying unit 330 increases the detection frequency 601a for the target imaging block corresponding to the target image block in the detection frequency map 600 by one. The detection frequency 601b of each imaging block around the imaging block of interest is increased by one. That is, the specifying unit 330 increases the detection frequency 601a of the imaging block (target imaging block) in which a moving object is detected in the plurality of imaging blocks, and increases the detection frequency 601b of each imaging block around the imaging block. . As a result, even if a moving object such as a person is not completely stationary but is moving finely, the detection frequency for the imaging block in which the moving object mainly exists and the imaging blocks around the imaging block The detection frequency can be increased. Therefore, it is possible to suppress registration of the image information of the moving body image acquired from the image block corresponding to the imaging block around the imaging block in the cache model in the background model 500. Therefore, the accuracy of moving object detection is further improved.

In the present modification, as illustrated in FIG. 23, in step s143, the specifying unit 330 increases the non-detection frequency 611a for the target imaging block corresponding to the target image block in the non-detection frequency map 610 by one. In addition, the non-detection frequency 611b of each imaging block around the imaging block of interest is increased by one. That is, the specifying unit 330 increases the non-detection frequency 611a of an imaging block (target imaging block) in which no moving object has been detected in the plurality of imaging blocks, and the non-detection frequency 611b of each imaging block around the imaging block. Increase. As a result, even if the moving object is not completely stationary and moves finely, it is possible to appropriately clear the detection frequency for an imaging block that is not currently likely to have a moving object. Therefore, the accuracy of moving object detection is further improved.

As described above, when the detection frequency map 600 is updated after the moving object detection process in step s12 is completed and before the background model update process in step s14 is started, the process is illustrated in FIG. As described above, the detection frequency 601c of the plurality of imaging blocks BKc respectively corresponding to the plurality of image blocks identified as moving body images in step s12 is increased by 1, and each of the imaging around the plurality of imaging blocks BKc is increased. The detection frequency 601d of the block BKd may be increased by one.

Similarly, when the non-detection frequency map 610 is updated after the moving object detection process in step s12 is completed and before the background model update process in step s14 is started, it is not a moving object image in step s12. The non-detection frequency of the plurality of imaging blocks respectively corresponding to the plurality of image blocks identified as “1” may be increased by one, and the non-detection frequency of each imaging block around the plurality of imaging blocks may be increased by one. .

In the above example, the size of the image block is 3 pixels × 3 pixels. However, the size is not limited to this, and the size of the image block may be 4 pixels × 4 pixels or 5 pixels × 5 pixels. Good.

In the above example, the code word CW for a certain image block exemplifies a case where the pixel values of all the pixels in the certain image block are included as image information. However, the present invention is not limited to this. The code word CW may not include the pixel values of all the pixels in the image block as the image information. Specifically, when the size of the image block is 3 pixels × 3 pixels, the code word CW may include pixel values for 5 pixels as image information. Thus, since the amount of processing can be reduced by reducing the amount of information in the code word CW, the moving object detection processing can be speeded up.

In the above example, it is assumed that each pixel in the input image 200 has a pixel value of R (red), G (green), and B (blue). It is not limited. Specifically, the pixel value of each pixel in the input image 200 may be expressed using a color space other than RGB. For example, when the input image 200 is YUV format image data, the luminance signal Y and the two color difference signals U and V are used as pixel values of each pixel.

As described above, the detection apparatus 1 has been described in detail, but the above description is an example in all aspects, and the present invention is not limited thereto. The various modifications described above can be applied in combination as long as they do not contradict each other. And it is understood that the countless modification which is not illustrated can be assumed without deviating from the scope of the present invention.

DESCRIPTION OF SYMBOLS 1 Detection apparatus 5 Background model memory | storage part 31 Moving body detection part 311 Control program 330 Specification part 331 Registration determination part

Claims

A storage unit for storing a background model including background image information;
Using the background model and the input image, a detection unit for detecting a detection target existing in an imaging region shown in the input image;
In the imaging region, a specifying unit that specifies a region where a detection target is likely to exist based on a detection result in the detection unit;
A detection device comprising: a determination unit that determines whether or not to register image information obtained from an input image as background image information in the background model using a specific result in the specifying unit.
The detection device according to claim 1,
The specifying unit obtains the detection frequency of the detection object for each of a plurality of partial imaging regions constituting the imaging region based on a detection result in the detection unit, and among the plurality of partial imaging regions, The detection apparatus which makes the partial imaging area whose detection frequency is more than a 1st threshold value or larger than the said 1st threshold value the area | region where possibility that the said detection target object exists is high.
The detection device according to claim 2,
The specifying unit obtains the non-detection frequency of the detection object for each of the plurality of partial imaging regions based on the detection result in the detection unit,
The specifying unit includes the detection frequency and the non-detection frequency for a partial imaging region in which the non-detection frequency is greater than or equal to a second threshold value or greater than the second threshold value among the plurality of partial imaging regions. Clear the detection device.
The detection device according to claim 2,
The specifying unit increases the detection frequency of the first partial imaging region in which the detection target is detected in the plurality of partial imaging regions, and includes the second partial imaging region around the first partial imaging region. A detection device for increasing the detection frequency.
The detection device according to claim 3,
The specifying unit increases the detection frequency of the first partial imaging region in which the detection target is detected in the plurality of partial imaging regions, and includes the second partial imaging region around the first partial imaging region. A detection device for increasing the detection frequency.
A detection method of a detection object in a detection device,
(A) using a background model including background image information and an input image, detecting a detection target existing in an imaging region shown in the input image;
(B) in the imaging region, a step of specifying a region where a detection target object is likely to exist based on a detection result in the step (a);
(C) detecting a detection target, comprising determining whether or not image information obtained from an input image is registered in the background model as background image information using the specific result in the step (b). Method.
A control program for controlling a detection device that detects a detection object,
In the detection device,
(A) using a background model including background image information and an input image, detecting a detection target existing in an imaging region shown in the input image;
(B) in the imaging region, a step of specifying a region where a detection target object is likely to exist based on a detection result in the step (a);
(C) A control program for executing a step of determining whether or not to register image information obtained from an input image as background image information in the background model using the specific result in the step (b).