JP6562648B2 - Image processing apparatus, image processing method, and program - Google Patents

Description

  The present invention relates to an image processing apparatus, an image processing method, and a program.

In recent years, the resolution of videos to be processed has been increased, and the processing amount per frame of video has increased in proportion to this. Therefore, an increase in processing amount is a problem particularly when high-speed processing such as real-time processing is required. On the other hand, there is generally known a method of speeding up processing by dividing each frame image of a video into regions and sharing the divided image processing units in units of partial region images to perform parallel processing.
However, when performing image recognition processing, an object that crosses the division boundary in the original image is divided at the division boundary, so that the image including the entire area of the object is not input to any image processing unit, so recognition fails. There was a problem.
In order to solve this problem, Japanese Patent Application Laid-Open No. H10-228707 detects a recognition target that cannot be detected in a divided image by generating a reduced image of the original image separately from the partial region image and processing the reduced image with an additional image processing unit. Is disclosed.

Japanese Patent No. 5235808

However, in the conventional technology, since the recognition target that crosses the image division boundary can be recognized only with the reduced image of the original image, the recognition accuracy of the recognition processing is higher than when the image recognition processing is performed using the image before the reduction. There was a problem of deterioration.
Therefore, the present invention provides an image processing method capable of performing recognition processing of a recognition target object related to the division boundary with higher accuracy when an image is divided for processing by a plurality of image processing units. For the purpose.

Therefore, the image processing apparatus of the present invention includes a switching means for switching the area dividing method of the input temporally continuous images, a plurality of divided partial region images were based on the switched that the area dividing method by the switching means, respectively anda supply means for supplying to the corresponding plurality of image recognition processing means, said switching means, for the first input image, by applying the first area division pattern, the first input for the next second input image of the image, sequential area dividing method of the first area division pattern and the input temporally continuous images by applying a different second area division pattern is switching said supply means, the cutting of despite an image at least two or more partial-area image is processed in the same image recognition processing means of the plurality of divided partial area images on the basis of the area dividing method If not, joining the divided partial area image as one image, and supplies to the image recognition processing unit.

  According to the present invention, when an image is divided for processing by a plurality of image processing units, recognition processing of a recognition target object related to the division boundary can be performed with higher accuracy.

It is a figure which shows an example of the hardware constitutions of an image recognition apparatus. 3 is a flowchart illustrating an example of information processing of the image recognition apparatus according to the first embodiment. It is a figure which shows an example of the data format of area | region division pattern information. It is a figure which shows an example which changes an area | region division | segmentation pattern for every flame | frame of a different time of an input image | video. It is a figure which shows an example of the determination method of an area | region division pattern. It is a figure which shows an example of the supply method to the image recognition process part of the partial area image by which area | region division | segmentation was carried out. It is a figure which shows an example of the read-out order of the image data supplied to an image recognition process part. 10 is a flowchart illustrating an example of information processing of the image recognition apparatus according to the second embodiment. It is a figure which shows an example of the data format of recognition information. It is a figure which shows an example of the processing operation which changes an area | region division pattern for every flame | frame using recognition information.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

<Embodiment 1>
FIG. 1 is a diagram illustrating an example of a hardware configuration of the image recognition apparatus. The image recognition device is an example of an image processing device.
The memory 100 is used for temporary storage of data. The memory 100 receives video data and area division pattern information input from the input / output interface 101, recognition information output from the first image recognition processing unit 106 and the second image recognition processing unit 107, and the like via the bus system 103. And saved.
The input / output interface 101 receives externally input video data and area division pattern information of the video data and outputs an image recognition result to the outside based on an instruction from the overall control unit 102. Video data is input sequentially and sequentially in time order for each frame. Input data is transmitted and stored in the memory 100 via the bus system 103.
The overall control unit 102 controls the processing order and processing contents of each processing unit.
The bus system 103 mediates transmission / reception of data between the processing units.
Based on an instruction from the overall control unit 102, the region division control unit 104 receives the region division pattern information stored in the memory 100 and the recognition information output from the first image recognition processing unit 106 and the second image recognition processing unit 107. Read out via the bus system 103. The area division control unit 104 determines two image areas to be extracted from the input video frame based on the area division pattern information or the recognition information, and notifies the partial area image supply unit 105 of the image area information.
The partial region image supply unit 105 reads out the input video frames stored in the memory 100 based on the image region information notified from the region division control unit 104 via the bus system 103, and performs the first image recognition processing unit 106, The image is supplied to the second image recognition processing unit 107. That is, the region division control unit 104 sequentially switches the region division method for temporally continuous input images (input video frames). Then, the partial region image supply unit 105 divides the input image according to the region division method, supplies one partial region image to the first image recognition processing unit 106, and supplies the other divided region image to the second image recognition processing unit 107. To supply.
The first image recognition processing unit 106 and the second image recognition processing unit 107 perform image recognition processing on the video frame (partial region image) supplied from the partial region image supply unit 105. Through the image recognition process, position information indicating the existence area of the recognition target in the input frame and identification information of the recognition target are transmitted to the memory 100 via the bus system 103 and stored.

In this configuration, there are two image recognition processing units, but a configuration having three or more image recognition processing units may be used. In that case, the region division control unit 104 determines a partial region of an image to be supplied to each image recognition processing unit according to the number of image recognition processing units, and notifies the partial region image supply unit. The partial area image supply unit 105 reads the image of the area notified from the area division control unit 104 from the memory 100 and supplies the image to the corresponding image recognition processing unit.
In FIG. 1, for example, the area division control unit 104 and the partial area image supply unit 105 have been described as hardware configurations. However, these may be implemented in the image recognition apparatus as a software configuration. That is, the functions of the region division control unit 104 and the partial region image supply unit 105 may be realized by the overall control unit 102 executing processing based on a program stored in the memory 100. Similarly, the overall control unit 102 may execute processes based on a program stored in the memory 100 to realize each process in the flowchart described below.

  FIG. 2 is a flowchart illustrating an example of information processing of the image recognition apparatus according to the first embodiment. In step S <b> 200, the overall control unit 102 stores the area division pattern information input via the input / output interface 101 in the memory 100 via the bus system 103. An example of the data format of the area division pattern information is shown in FIG. The area division pattern information includes a first area division pattern 300 and a second area division pattern 301 shown in FIG. Further, the area division patterns 300 and 301 are composed of a start point x coordinate 302 indicating the upper left corner coordinates of the rectangular area, a start point y coordinate 303, an end point x coordinate 304 indicating the lower right end of the rectangular area, and an end point y coordinate 305. In the present embodiment, one area division pattern corresponds to rectangular area information including a start point x coordinate 302, a start point y coordinate 303, an end point x coordinate 304, and an end point y coordinate 305. May correspond to the rectangular area information. Further, although there are two area division patterns in the present embodiment, they may have three or more.

FIG. 4A is a diagram illustrating an example of a division state when region division is performed based on the first region division pattern 300. The first input image 400 in FIG. 4A is the entire frame. The first area division pattern first image recognition processing area 402 is an image area that is processed by the first image recognition processing unit 106, and is within a rectangular area indicated by the coordinate information 302 to 305 of the first area division pattern 301. The first area division pattern second image recognition processing area 403 is an image area processed by the second image recognition processing unit 107, and is an outer rectangle excluding the rectangular areas indicated by the coordinate information 302 to 305 of the first area division pattern 301. It is an area.
On the other hand, FIG. 4B is a diagram showing an example of a division state when the area division is performed based on the second area division pattern 301. In this example, the second area division pattern first image recognition processing area 404 is within the rectangular area indicated by the coordinate information of the second area division pattern 301. The left-side rectangular area and the right-side rectangular area outside the second area division pattern first image recognition processing area 404 are the second area division pattern second image recognition processing area 405. As in this example, depending on the region division pattern, the second image recognition processing region is two image regions that are spatially divided. In the configuration according to the present embodiment, the image areas to be given to the inputs of the two image recognition processing units are determined in the form of the rectangular area and the rectangular area outside the range by the coordinate data indicating one rectangular area. However, this partial area determination method is an example, and another method may be used.

An example of the method for determining the region division pattern is shown in FIG. As shown in FIG. 5, the reference area division pattern 500 equally divides the input image into two in the left-right direction with the width w. Partial-area image supply unit 105, a reference area division pattern first image recognition processing region 50 3 the first image recognition processing unit 106, a reference area division pattern the second image recognition processing region 502 second image recognition obtained by dividing This is supplied to the processing unit 107. W / 2 division boundary with respect to the reference area dividing pattern 500 in the second region division pattern 50 4 contrary, that only half of the width, are shifted. Thus, the position where there is division boundary in the reference area dividing pattern 500, in the second area division pattern first image recognition processing area 50 5 of the second region division pattern 50 4 becomes the center position. Thus, it is possible to capture the target object located in the division boundary in the reference area division pattern 500 to the center position of the second region division pattern 50 4 In recognition processing image. The center of the image area for the three image areas after division as the second area division pattern first image recognition processing area 50 5 in the second region division pattern 50 4, first image recognition processing unit 106 performs processing. Further, the left end, the second image recognition processing unit 107 performs processing the right edge of the area as the second area division pattern the second image recognition processing zone 50 6. The division pattern of this method can be similarly applied to the case of dividing an image in the vertical direction instead of the horizontal direction, and also in the case of dividing the image in both the horizontal direction and the vertical direction. Further, the number of divisions in the width direction or the height direction may be a value larger than two.
In the present embodiment, it is assumed that when the region division pattern is changed, the first image recognition processing unit 106 and the second image recognition processing unit 107 determine the region division pattern so as to process partial region images having the same resolution. Yes. However, it may be a region division pattern in which the resolution of the partial region image processed by one image recognition processing unit is larger than the resolution of the partial region image processed by the other image recognition processing unit.

Returning to the description of FIG.
In step S <b> 201, the overall control unit 102 stores the frame data for one frame input via the input / output interface 101 in the memory 100 via the bus system 103.
In step S202, the area division control unit 104 switches area division patterns. The area division control unit 104 selects one of the area division pattern information stored in the memory 100 in step S200, reads it, and transmits it to the partial area image supply unit 105. The area division control unit 104 determines and selects area division pattern information so that the first area division pattern 300 and the second area division pattern 301 are alternately switched for each frame process. That is, the region division control unit 104 switches the divided region method of the input image that is temporally continuous. In the present embodiment, the region is divided in the left-right direction, but other division patterns such as division in the up-down direction may be taken. Further, when the number of area division patterns is 3 or more, selection may be made so that these patterns are cyclically switched sequentially. However, any other method may be used for selecting the area division pattern.
In step S <b> 203, the partial area image supply unit 105 generates and issues a read address to the memory 100 based on the notification of the area division pattern received from the area division control unit 104. In the example shown in FIG. 4A, the partial area image supply unit 105 reads the read address of the first area division pattern first image recognition processing area 402 and the read address of the first area division pattern second image recognition processing area 403. Are issued in parallel. Then, the partial area image supply unit 105 reads out from the memory 100 in parallel. The reading order of the image data may be the raster order, but other methods may be used. The partial area image supply unit 105 sequentially supplies the read image data to the first image recognition processing unit 106 and the second image recognition processing unit 107, respectively. That is, the partial region image supply unit 105 supplies a plurality of partial region images divided based on the region division method switched by the region division control unit 104 to a plurality of corresponding image processing units.

In the example of the division pattern of FIG. 4B, the partial area image supply unit 105 controls the reading order from the memory 100 for the three image areas as shown in FIG. The images are supplied to the first image recognition processing unit 106 and the second image recognition processing unit 107 one by one. A method of consolidating three regions into two will be described with reference to FIG. In FIG. 6, the rectangular area indicated by the coordinate information 302 to 305 of the second area division pattern 301 is defined as area B601, the left rectangular area outside this rectangular area is defined as area A600, and the right rectangular area is defined as area C602. The partial region image supply unit 105 supplies the image of the region B601 as it is to the first image recognition processing unit 106 as the first image recognition processing unit supply image 603. The partial area image supply unit 105 controls the reading order from the memory for the area A600 and the area C602 as one image of the format shown in the second image recognition processing unit supply image 604 obtained by joining the two images. The images are supplied to the second image recognition processing unit 107 in raster order.
Image reading order instruction lines 700 to 705 in FIG. 7 indicate the reading order of the image data. As shown in FIG. 7, the partial area image supply unit 105 reads first the first line of the area A600, second reads the first line of the area C602, secondly reads the second line of the area A600, Third, readout of the second line in the region C602 and pixel values are read out. That is, the partial area image supply unit 105 alternately reads out the area A600 and the area C602 in order from the first line. The order instruction line 704 and the order instruction line 705 are read pixel final lines of the area A600 and the area C602, respectively.
By controlling in this way, the second image recognition processing unit 107 processes only two areas B601, such as an area A600 and an area C602, or only an area B601, according to an area division pattern. Can be treated as a single image. This facilitates control of the image recognition processing unit.

However, the junction boundary between the area A and the area C in the second image recognition processing unit supply image 604 is not an area that is originally spatially adjacent. Therefore, the partial region image supply unit 105 notifies the second image recognition processing unit 107 of the joint position information together with the supply of the second image recognition processing unit supply image 604. Then, when the second image recognition processing unit 107 detects a region that crosses the notified joint boundary as a recognition target, the second image recognition processing unit 107 performs control so that it is regarded as a false detection and is not transmitted to the memory 100, that is, the processing result is discarded. .
In step S204, the first image recognition processing unit 106 and the second image recognition processing unit 107 perform image recognition processing on the input partial region images, and generate recognition information for each detection region. The first image recognition processing unit 106 and the second image recognition processing unit 107 write the processing result to the memory 100 via the bus system 103. The process of step S204 is an example of a process result storage process.
After completing the process in step S204, the overall control unit 102 returns the process to step S201 and accepts the frame data of the next frame. Thereafter, the image recognition apparatus repeats the processing from step S201 to step S204.

FIG. 4A and FIG. 4B are diagrams illustrating an example of a state in which different region dividing methods are applied to frames at different times. A first input image 400 in FIG. 4A is an entire image of a frame at a certain time t. In the second input image 406, the recognition target 401 exists across the boundary between the first region division pattern first image recognition processing region 402 and the first region division pattern second image recognition processing region 403. Therefore, since the input images of the first image recognition processing unit 106 and the second image recognition processing unit 107 include only a part of the recognition target 401, the image recognition process is difficult.
FIG. 4B shows an example of a state in which a region division pattern different from that in FIG. 4A is applied to the frame at time t + Δt immediately after FIG. The second input image 406 is an entire image of the frame at time t + Δt. In FIG. 4B, since the recognition target 401 is within the second region division pattern first image recognition processing region 404, the first image recognition processing unit 106 that receives this image receives the image recognition of the recognition target 401. Easy to process.
By switching the region division method for each frame with a sufficiently short time interval, recognition targets that fail to be recognized in a certain frame can be recognized in different frames that are close in time. To avoid problems. In addition, since the image is not reduced as in the method described in Patent Document 1, image recognition processing can be performed using an image with the original resolution, and thus higher image recognition accuracy can be obtained.

<Embodiment 2>
The hardware configuration of the image recognition apparatus of the second embodiment is the same as that of the first embodiment as shown in FIG. However, in the second embodiment, the functions of the area division control unit 104 are different. The contents will be described later.
FIG. 8 is a flowchart illustrating an example of information processing of the image recognition apparatus according to the second embodiment. In FIG. 8, steps S200, S201, S202, S203, and S204 are the same as those in the flowchart of FIG. In step S <b> 800, the region division control unit 104 reads the recognition information that is the result output from the first image recognition processing unit 106 and the second image recognition processing unit 107 stored in the memory 100.
An example of the data format of the recognition information output from the first image recognition processing unit 106 and the second image recognition processing unit 107 is shown in FIG. In FIG. 9, a recognition target area 900 indicates a rectangular area where an object detected by image recognition processing exists, and further indicates a start point x coordinate 902, a start point y coordinate 903, and a lower right end coordinate indicating the coordinates of the upper left end of the rectangle. It consists of an end point x coordinate 904 and an end point y coordinate 905. The recognition target attribute 901 is information indicating the characteristics of the recognition target. In this embodiment, the recognition target is a person, and the recognition target attribute 901 includes information on gender 906, age 907, and clothing 908. The recognition target may be other than a person, and does not limit the present embodiment. The clothes 908 may be a set of a plurality of information such as 1-bit information indicating whether or not glasses are worn and 1-bit information indicating whether or not masks are being worn. Further, a plurality of pieces of recognition information data may be stored in the memory 100 according to the number of recognition targets detected with the format of FIG. 9 as one set.

The area division control unit 104 refers to the recognition information read from the memory 100 and determines whether there is a recognition target to be tracked. The area division control unit 104 compares the recognition target attribute information set as a tracking target in advance with the recognition information of the recognition target read from the memory 100, and checks whether the difference of the attribute information is within a certain range. Thus, it is determined whether a certain recognition target is a tracking recognition target. If the tracking recognition target exists in the stored recognition information, the region division control unit 104 proceeds to step S801. If not, the region division control unit 104 proceeds to step S202.
In step S801, the region division control unit 104 estimates an existing region in the next frame to be tracked. The area division control unit 104 performs the estimation using recognition information for several past frames. First, the area division control unit 104 extracts recognition information to be tracked and compares attribute information between the recognition information of each frame. The area division control unit 104 regards recognition targets whose differences are within a certain range as the same recognition target. The area division control unit 104 linearly approximates the coordinate data 902 to 905 of each recognition target area 900 in each frame regarded as the same in the time direction, and sets the value of the next frame time of each approximate line to the next frame. Assume that the time coordinate data 902 to 905 are estimated values. The area division control unit 104 sets the rectangular area indicated by these as the recognition target estimation area. The method for estimating the region in the next frame of the recognition target region is an example, and other methods may be used.

In step S802, the region division control unit 104 determines a region division method for the next frame using the tracking recognition target region information obtained in step S801. The determination method will be described with reference to FIG. FIG. 10A shows a state where the time t input frame 1000 is divided into a first image recognition processing unit processing region 1001 and a second image recognition processing unit processing region 1002. Here, the time t recognition target area 1003 is an area where the first image recognition processing unit 106 detects the person of the recognition target 1005 in the first image recognition processing unit processing area 1001. The first image recognition processing unit 106 detects recognition target areas for a plurality of past frames including the time t input frame 1000 and stores the recognition information in the memory 100. The area division control unit 104 recognizes the time t + Δt frame recognition target area estimated from the time t recognition target area 1003 and the recognition target area of the recognition information of the frame before time t as the time t estimated recognition target area 1004. As shown in FIG. 10A, when the time t estimation recognition target area 1004 extends over the first image recognition processing section processing area 1001 and the second image recognition processing section processing area 1002, the area division control section 104 determines that the time t + Δt The region division pattern is determined so that the region division boundary is not located within the region of the time t estimation recognition target region 1004 in the input frame 1010.
FIG. 10B is a diagram showing a state where the area division pattern is changed by the above method. The time t + Δt input frame 1010 is divided into a first image recognition processing unit processing area 1011 and a second image recognition processing unit processing area 1012 for processing. If the position of the recognition target 1005 is the region division method of the time t input frame 1000, the region division boundary is crossed in the time t + Δt input frame 1010. However, since the region division method is changed so that the region division boundary is not located in the time t estimation recognition target region 1004 in step S801, the entire recognition target 1005 is included in the first image recognition processing unit processing region 1011. include. Therefore, the first image recognition processing unit 106 can detect the recognition target 1005. Thus, it is possible to control the tracking target so as not to cross the region division boundary by changing the region division pattern based on the movement prediction of the recognition region of the tracking target.

The area division control unit 104 processes information about the time t + Δt recognition area 1013, which is a detection area when the first image recognition processing unit 106 processes the time t + Δt input frame 1010 and detects the recognition target 1005, and before the time t + Δt. The time t + Δt estimation recognition target area 1014 in the next time frame is obtained from the recognition information of the current frame. Then, by repeating the above processing, a recognition target close to the feature of the recognition target attribute information set in advance can be reliably recognized in each frame.
Here, in step S800, when there are a plurality of recognition targets whose difference from the recognition target attribute set in advance in the recognition target attribute 901 is a predetermined value or less, the region division control unit 104 selects a recognition target region having a smaller difference value. The region division pattern may be controlled so that it is not preferentially arranged on the region division boundary. By doing in this way, it is possible to give priority to an object closer to a desired attribute and avoid being located on the region division boundary.
In the process according to the present embodiment, if there is no tracking recognition target in step S800, the process proceeds to step S202, and the area division pattern stored in advance in the memory 100 is applied. Therefore, when there is no tracking recognition target, the partial region image supply unit 105 is switched to the first image recognition processing unit 106 and the second image recognition processing unit 107 while switching the region division pattern set by performing the same operation as in the first embodiment. Will supply partial area images. When the tracking recognition target appears, steps S800 to S802 are executed, and the division pattern is controlled so that the tracking recognition target region is not always located on the region division boundary.
By controlling as described above, if there is no target to be focused on, operate so that the entire area of the input video can be recognized evenly, and if the target to be focused on appears in the input video The operation can be performed so as to further increase the recognition accuracy of the target recognition target.

<Other embodiments>
The present invention supplies a program that realizes one or more functions of the above-described embodiments to a system or apparatus via a network or a storage medium. It can also be realized by a process in which one or more processors in the computer of the system or apparatus read and execute the program. It can also be realized by a circuit (for example, ASIC) that realizes one or more functions.

  As described above, according to the processing of each embodiment described above, it is possible to perform image recognition processing of objects in all regions in an image without degrading the recognition accuracy of image recognition.

100 Memory 101 Input / Output Interface 102 Overall Control Unit 103 Bus System 104 Region Division Control Unit 105 Partial Area Image Supply Unit 106 First Image Recognition Processing Unit 107 Second Image Recognition Processing Unit

Claims (10)

A switching means for switching a region dividing method of temporally continuous input images;
A supply means for supplying a plurality of image recognition processing means for a plurality of partial region images, corresponding respectively divided based on switched that the area dividing method by the switching means,
Have
It said switching means, for the first input image, by applying the first area division pattern, for the next second input image of the first input image, the first region divided sequentially switching the area dividing method of the input temporally continuous images by applying a different second area division pattern is a pattern,
The supply means is divided even though at least two of the partial area images divided based on the area dividing method are images processed by the same image recognition processing means. In this case, an image processing apparatus that joins the divided partial area images into one image and supplies the image to the image recognition processing means . A switching means for switching a region dividing method of temporally continuous input images;
A supply means for supplying a plurality of image recognition processing means for a plurality of partial region images, corresponding respectively divided based on switched that the area dividing method by the switching means,
Have
It said switching means, for the first input image, by applying the first area division pattern, for the next second input image of the first input image, the first region divided sequentially switching the area dividing method of the input temporally continuous images by applying a different second area division pattern is a pattern,
The switching means determines whether or not a tracking recognition target exists based on the recognition information output by the plurality of image recognition processing means, and determines that the tracking recognition target exists. An image processing apparatus that estimates a position and determines a region division method of the input image so that a boundary in region division is not located at the estimated position . The first area division pattern is a region division pattern for dividing a continuous area of width equal to the direction in an input image,
The second area division pattern, the division position of a region division pattern for dividing the input image in a position where only shifted along the direction half the width of the width Claim 1 image processing apparatus according . The image processing apparatus of the plurality of image recognition processing according to claim 1 to 3 further comprising means any preceding claim. Each of the plurality of image recognition processing means performs an image recognition process on the supplied partial area image, and uses position information indicating the existence area of the recognition target and attribute information indicating the attribute of the recognition target as a recognition result. the image processing apparatus according to claim 1 to 4 any one of claims outputs. Wherein the plurality of image recognition processing unit, when obtaining the region spanning the junction boundary of the bonded portion area image as a recognition information, the recognition information discarding claim 1 image processing apparatus according. Further comprising processing result storage means for storing the recognition information output by the plurality of image recognition processing means in a memory;
The switching means determines whether or not there is a tracking recognition target based on the recognition information stored in the memory by the processing result storage means, and determines that there is no tracking recognition target. The image processing apparatus according to claim 1, wherein a region dividing method for successive input images is sequentially switched. An image processing method executed by an image processing apparatus,
A switching step for switching a region dividing method of temporally continuous input images;
A plurality of divided partial region images were based on the switched that the area dividing method by the switching step, a supply step of supplying to the corresponding plurality of image recognition processing step,
Including
In the switching step, for the first input image, by applying the first area division pattern, to the first input image following the second input image, the first region divided sequentially switching the area dividing method of the input temporally continuous images by applying a different second area division pattern is a pattern,
In the supplying step, at least two or more partial region images among the plurality of partial region images divided based on the region dividing method are divided although they are images processed in the same image recognition processing step. In this case, an image processing method in which the divided partial area images are joined to form one image and supplied to the image recognition processing step . An image processing method executed by an image processing apparatus,
A switching step for switching a region dividing method of temporally continuous input images;
Supplying a plurality of partial region images divided based on the region dividing method switched by the switching step to a plurality of corresponding image recognition processing steps,
Including
In the switching step, the first area division pattern is applied to the first input image, and the first area division is applied to the second input image next to the first input image. By applying a second region segmentation pattern different from the pattern, the region segmentation method of the temporally continuous input image is sequentially switched,
In the switching step, based on the recognition information output by the plurality of image recognition processing steps, it is determined whether there is a tracking recognition target, and if it is determined that the tracking recognition target exists, the tracking recognition target An image processing method for estimating a position and determining a region division method of the input image so that a boundary in region division is not located at the estimated position. A program for causing a computer to function as each unit of the image processing apparatus according to any one of claims 1 to 7 .

Images