JP2015219712A - Same object detection device, same object detection method, and same object detection program - Google Patents

Abstract

An identical object detection device that suppresses erroneous detection of the same person is provided. A target detection unit for detecting a first determination target and a second determination target from a first captured image captured in time series and a second captured image captured in time series; When the first determination target and the second determination target are each divided into a plurality of areas, and the color information of the corresponding areas is compared, the change of the color information in the time series of the comparison areas Whether the first determination target and the second determination target are the same by using a second weighting factor that is smaller than the first weighting factor used for any one of the regions larger than that region The same object detection device 400 having a difference determination unit 470 for determining whether or not. [Selection] Figure 2

Description

  This case relates to the same object detection device, the same object detection method, and the same object detection program.

  A system for detecting the same person using images from a plurality of cameras whose monitoring ranges do not overlap is known (see, for example, Patent Documents 1 to 5). For example, an image acquired from a video is divided into a plurality of regions, and feature quantities of each region are extracted using a color histogram, so that the target person can be identified without depending on the change in the posture of the person included in the video. There is a technique for searching (see, for example, Patent Document 6).

JP 2005-202938 A JP 2007-11696 A Japanese Patent Laid-Open No. 11-96364 Japanese Patent Laid-Open No. 2005-250692 JP 2009-231921 A JP 2011-18238 A

  However, although the above-described technique does not depend on changes in posture within each region, if the posture changes across regions, the target person search accuracy decreases. In this case, the same person may be detected as a different person.

  Therefore, in one aspect, an object of the present invention is to provide an identical object detection device, an identical object detection method, and an identical object detection program that suppress erroneous detection of the same person.

  The same object detection device disclosed in the present specification determines a first determination object and a second determination object from each of a first captured image captured in time series and a second captured image captured in time series. When the object detection unit to detect, the first determination object, and the second determination object are each divided into a plurality of areas and the color information of the corresponding areas is compared, The second weighting factor smaller than the first weighting factor used for any one of the regions is used for a region where the change in the color information in the series is larger than any one of the regions, And the same determination unit that determines whether or not the second determination target is the same.

  The same target detection method disclosed in this specification is configured to determine a first determination target and a second determination target from each of a first captured image captured in time series and a second captured image captured in time series. When the first judgment object and the second judgment object are each divided into a plurality of areas and the color information of the corresponding areas is compared, the time series of the areas to be compared is detected. By using a second weighting factor that is smaller than the first weighting factor used for any one of the regions in a region where the change in color information is larger than any one of the regions, the first determination target The same object detection method in which the computer executes a process of determining whether or not the second determination objects are the same.

  The same target detection program disclosed in the present specification is configured to determine a first determination target and a second determination target from each of a first captured image captured in time series and a second captured image captured in time series. When the first judgment object and the second judgment object are each divided into a plurality of areas and the color information of the corresponding areas is compared, the time series of the areas to be compared is detected. By using a second weighting factor that is smaller than the first weighting factor used for any one of the regions in a region where the change in color information is larger than any one of the regions, the first determination target It is the same object detection program for making a computer perform a process which judges whether the said 2nd judgment object is the same.

  According to the same object detection device, the same object detection method, and the same object detection program that suppress erroneous detection of the same person disclosed in this specification, erroneous detection of the same object can be suppressed.

FIG. 1 is a diagram for explaining an example of the same object detection system. FIG. 2 is an example of a block diagram of the same object detection apparatus. FIG. 3 is an example of a hardware configuration of the same object detection apparatus. FIG. 4 is a flowchart illustrating an example of person information registration processing executed by the same target detection apparatus. FIG. 5 is an example of a color table. FIG. 6 is a diagram for explaining an example of a change in posture of a person image. FIG. 7 is a diagram for explaining an example of the orientation of a person image. FIG. 8 is an example of a color table in which weighting factors are registered. FIG. 9 is an example of a person table. FIG. 10 is a flowchart illustrating an example of the color table update process. FIG. 11 is a flowchart illustrating an example of the association process executed by the same target detection apparatus. FIG. 12 is a flowchart illustrating an example of similarity calculation processing. FIG. 13 is a diagram for explaining an example of similarity calculation processing.

  Hereinafter, an embodiment for carrying out this case will be described with reference to the drawings.

  FIG. 1 is a diagram for explaining an example of the same object detection system S. As shown in FIG. 1, the same target detection system S includes a plurality of imaging devices 100, 200, 300 and the same target detection device 400.

  The imaging devices 100, 200, and 300 each take an image within its own imaging range. Examples of the imaging devices 100, 200, and 300 include devices that capture moving images and videos such as video cameras. Each imaging device 100, 200, 300 is provided so that the imaging range does not overlap with the wall or ceiling of the passage through which the person M passes. Accordingly, in FIG. 1, the imaging device 100 captures an image within the imaging range including the clock 10 hung on the wall. The imaging device 200 captures an image within the imaging range including the painting 20 hung on the wall. Here, when the person M holding the bag B enters the imaging range of the imaging device 100, an image including the person M holding the bag B and the clock 10 is input to the imaging device 100. A part of the video includes an instantaneous captured image 30 at a certain time. In addition, when the person M moves straight from the imaging range of the imaging apparatus 100 and turns left in the passage, and enters the imaging range of the imaging apparatus 200, an image including the person M and the painting 20 is input to the imaging apparatus 200. Part of the video includes an instantaneous captured image 31 at another time. Since the imaging device 100 continuously captures images, a time-series captured image including the captured image 30 is input to the imaging device 100 as a video. The same applies to the imaging apparatus 200.

  The same object detection device 400 is connected to each imaging device 100, 200, 300. An example of the same object detection device 400 is a server device. The same target detection device 400 acquires each video input to the imaging devices 100, 200, and 300, and determines whether a person included in one of the videos is the same as a person included in another video. Note that the determination target by the same target detection device 400 is not limited to a person, but may be a moving body such as an animal or a robot that moves while moving a part of the body and changes its posture.

  Next, with reference to FIG. 2, a schematic functional configuration of the above-described same object detection device 400 will be described.

FIG. 2 is an example of a block diagram of the same object detection device 400.
As shown in FIG. 2, the same target detection device 400 includes a video acquisition unit 410, a target detection unit 420, a color information generation unit 430, an information registration unit 440, and an information storage unit 450. Further, the same target detection device 400 includes an information extraction unit 460, a difference determination unit 470, and an information update unit 480. Note that the information storage unit 450 may be outside the same object detection device 400.

The video acquisition unit 410 acquires video input to the imaging devices 100, 200, and 300. The video acquisition unit 410 converts the acquired video into an image for each frame. Thereby, a plurality of time-series images (captured images) by the imaging device 100 and the imaging device 200 are generated. The video acquisition unit 410 outputs a plurality of time-series images to the target detection unit 420.
The target detection unit 420 detects a person as a determination target for each image from a plurality of time-series images. Thereby, a plurality of persons whose postures change with the passage of time are detected. After detecting a plurality of persons, the target detection unit 420 tracks the same person between temporally continuous images. The target detection unit 420 outputs the tracked person to the color information generation unit 430.

The color information generation unit 430 extracts color information from the person tracked by the target detection unit 420. The color information generation unit 430 generates a color table by updating the color table using the extracted color information. The color information generation unit 430 outputs the generated color table to the information registration unit 440.
The information registration unit 440 includes the person feature amount of the person characterized by the color table in the person information, and registers the person information in the information storage unit 450. As a result, the personal information is stored in the information storage unit 450.

The information extraction unit 460 extracts the comparison source person information and the comparison destination person information from the information storage unit 450. The information extraction unit 460 outputs the extracted comparison source person information and comparison target person information to the difference determination unit 470.
The difference determination unit 470 compares the person information of the comparison source with the person information of the comparison destination and determines whether or not the person information is the same. The difference determination unit 470 outputs the determination result to the information update unit 480.

  The information update unit 480 updates the comparison source personal information stored in the information storage unit 450 according to the determination result. For example, if it is determined that the comparison-source person information and the comparison-destination person information are the same, the information update unit 480 uses the same label as the label assigned to the comparison-destination person information. To grant. Conversely, if the information update unit 480 determines that the comparison-source person information and the comparison-destination person information are different, the information update unit 480 uses a label different from the label assigned to the comparison-destination person information as the comparison-source person information. Give. As a result, the information storage unit 450 stores personal information with various labels. For example, if the same label is given to two pieces of person information, it can be determined that the persons specified by the pieces of person information are the same person.

  Next, the hardware configuration of the same object detection device 400 will be described with reference to FIG.

FIG. 3 is an example of a hardware configuration of the same object detection device 400.
As shown in FIG. 3, the same object detection device 400 includes at least a central processing unit (CPU) 400A, a random access memory (RAM) 400B, a read only memory (ROM) 400C, and a network I / F (interface) 400D. Yes. The same object detection device 400 may include at least one of a hard disk drive (HDD) 400E, an input I / F 400F, an output I / F 400G, an input / output I / F 400H, and a drive device 400I as necessary. The CPU 400A,..., The drive device 400I are connected to each other by an internal bus 400J. At least the CPU 400A and the RAM 400B cooperate to realize a computer.

An input device 510 is connected to the input I / F 400F. Examples of the input device 510 include a keyboard and a mouse.
A display device 520 is connected to the output I / F 400G. An example of the display device 520 is a liquid crystal display.
A semiconductor memory 530 is connected to the input / output I / F 400H. Examples of the semiconductor memory 530 include a universal serial bus (USB) memory and a flash memory. The input / output I / F 400H reads a program and data stored in the semiconductor memory 530.
The input I / F 400F and the input / output I / F 400H include, for example, a USB port. The output I / F 400G includes a display port, for example.

A portable recording medium 540 is inserted into the drive device 400I. Examples of the portable recording medium 540 include a removable disk such as a Compact Disc (CD) -ROM and a Digital Versatile Disc (DVD). The drive device 400I reads a program and data recorded on the portable recording medium 540.
The network I / F 400D includes, for example, a port and a physical layer chip (PHY chip). The same object detection device 400 is connected to a network via a network I / F 400D.

  In the RAM 400B described above, the program stored in the ROM 400C or the HDD 400E is stored by the CPU 400A. In the RAM 400B, the program recorded on the portable recording medium 540 is stored by the CPU 400A. When the stored program is executed by the CPU 400A, various operations described later are executed. In addition, what is necessary is just to make a program according to the flowchart mentioned later.

  Next, the operation of the same object detection device 400 will be described with reference to FIGS.

  FIG. 4 is a flowchart illustrating an example of person information registration processing executed by the same target detection device 400. FIG. 5 is an example of a color table. FIG. 6 is a diagram for explaining an example of a change in posture of a person image. FIG. 7 is a diagram for explaining an example of the orientation of a person image. FIG. 8 is an example of a color table in which weighting factors are registered. FIG. 9 is an example of a person table.

  First, the video acquisition unit 410 acquires a video input to any of the imaging devices 100, 200, and 300, and converts the acquired video into an image for each frame (step S101). For example, when a video input to the imaging apparatus 200 is acquired, a plurality of time-series images are generated by the imaging apparatus 200. Next, the target detection unit 420 detects a person from each image and extracts each person feature amount (step S102). More specifically, the target detection unit 420 detects a person as a determination target for each image from a plurality of time-series images using a predetermined detection method. As a detection method, for example, there are a background difference method, a dynamic binarization method, template matching, and the like (see JP 2002-157599 A). When the target detection unit 420 detects a person, the target detection unit 420 extracts a person feature amount representing the feature amount of the detected person. Examples of the person feature amount include a color at a detection position and a movement vector corresponding to a person who is temporally continuous.

  Next, the target detection unit 420 tracks the same person (step S103). More specifically, the target detection unit 420 is based on the person feature amount detected from the previous image (processed image) and the person feature amount detected from the current image (processed image) based on the determination criterion. The same person is tracked by associating. As a result, a moving person is detected and tracked in each image input to the imaging device 100 or the imaging device 200. The following technique may be used for the person tracking method. The shape change (appearance change) of the tracking target can be tracked by a technique using a color histogram. With regard to the movement (motion change) of an object, the region where the tracking target exists can be tracked by a technique that introduces a particle filter that samples based on the likelihood (Takayoshi Yamashita, Hironobu Fujiyoshi, “ Collaborative Object Tracking Using Relationships ”, Image Recognition Symposium (MIRU2011), pp. 56-63, July 2011).

Next, the color information generation unit 430 executes a color table update process (step S104). The color table update process is a process for extracting color information representing color characteristics from the person being tracked and writing the extracted color information in a color table having a predetermined format for updating. As a result, a color table including color information is generated. The color table is a table that summarizes what colors each region has when a human image representing a human image is divided into a plurality of regions (for example, rectangular regions). The area may be a pixel unit or a plurality of pixels. As shown in FIG. 5, the color table includes, as constituent elements, an area position that represents the position of each area, color information for each time in each area, and a weighting coefficient that will be described later. As shown in FIG. 6, the area position represents the position of each area when the person image is divided into a plurality of areas. The top (0, 0) of the region position is virtually set at the upper left corner. For example, the color information field of the area position (80, 10) at time _{t 0,} as shown in FIG. 5, the color information representing the color of the skin (220,165,110) are registered. The color information can be represented by, for example, a representative value in a region or a histogram.

The color table is determined for each direction of the person. For example, as shown in FIG. 7, when the orientation of a person is divided into four directions, the 0 ° direction facing the right direction, the 90 ° direction facing the front direction, the 180 ° direction facing the left direction, and the 270 ° direction facing the back direction. The four color tables are defined. In FIG. 5 described above, a color table in the 0 ° direction is shown. In place of such 90 ° increments, 45 ° increments may be used. In this case, eight color tables are defined. Although details of the color table update process will be described later, a color table including color information at time t ₀ is generated for each person's direction by the color table update process.

When the color table update process ends, the color information generation unit 430 determines whether or not the tracking ends (step S105). For example, when the person M goes out of the imaging range of the imaging apparatus 200 in FIG. 1, the person M is not input to the imaging apparatus 200, so the color information generation unit 430 determines that the tracking is finished (step S105: YES). On the other hand, when the person M is not out of the imaging range of the imaging apparatus 200, the person M is input to the imaging apparatus 200, and thus the color information generation unit 430 determines that the tracking is not finished (step S105: NO). . As long as the color information generation unit 430 determines that the tracking is not finished, the video acquisition unit 410 acquires the video. As a result, as shown in FIG. 5, the color information at times t ₀ , t ₁ ,..., T ₆ is extracted in order and registered in the corresponding color information column in the color table.

  If the color information generation unit 430 determines that the tracking is finished, then the information registration unit 440 calculates a weighting coefficient (step S106). The weighting coefficient is a coefficient representing the reliability of the similarity when the similarity is determined by comparing the same regions in different color tables. The smaller the weighting coefficient, the smaller the influence on the difference determination of the entire person image, and the easier it is to determine that they are the same in the difference determination of the entire person image. The weighting coefficient is calculated based on time-series color information in each region.

For example, as shown in FIG. 6, in the region position (80, 10) including a part of the face of the person M, as shown in FIG. 5, the color information (220, 165, 110) from time t ₀ to time t ₆ is shown. ) Has not changed. That is, even if the posture changes due to the movement of the arm, the color information is unlikely to change, and since the reliability is high when obtaining the similarity, a relatively large weighting factor is calculated. Conversely, as shown in FIG. 6, the area position including a portion of the legs of the person M (50, 100), as shown in FIG. 5, the color information from the time _{t 0} to time _{t 6} (168, 168, 168) and color information (1, 63, 101). That is, when the position of the eyelid B changes due to the movement of the arm, the color information (168, 168, 168) representing the color of the eyelid B and the color information representing the color of the clothing of the person M are displayed at the region position (50, 100). Since (1, 63, 101) appears, the color information is an easily changing region, and since the reliability when obtaining the similarity is low, a relatively small weight coefficient is calculated. In the region position including a portion of the arm of the person M (40, 70), as shown in FIG. 5, the color information from the time _{t 0} to time _{t 6} (1,63,101) and color information (60, 60, 60), but the color information does not change as much as the region position (50, 100) because it is affected by the shadow of the arm. Therefore, a weighting factor that belongs between the weighting factor calculated for the region position (80, 10) and the weighting factor calculated for the region position (50, 100) is calculated.

  As a specific calculation method of the weighting factor, for example, there is a method of plotting color information on a three-dimensional graph of the RGB axis and calculating the weighting factor by examining the distribution and dispersion of the color information. When the distribution or variance is small, the color information does not change much even if the posture changes, and therefore the information registration unit 440 calculates a large weighting factor according to a predetermined small distribution. Conversely, when the distribution or variance is large, various color information appears when the posture changes, and the color information changes greatly. Therefore, the information registration unit 440 calculates a small weighting factor according to a predetermined large distribution. To do.

  As a result, as shown in FIG. 8, the information registration unit 440 registers the weighting factor “2.0” calculated for the region position (80, 10) in the corresponding weighting factor column in the color table. Similarly, the information registration unit 440 registers the weighting factor “1.0” calculated for the region position (40, 70) in the corresponding weighting factor column in the color table. The information registration unit 440 registers the weighting factor “0.3” calculated for the region position (50, 100) in the corresponding weighting factor column in the color table.

  After calculating the weighting factor, the information registration unit 440 then registers person information (step S107). As shown in FIG. 9, the person information includes a person ID for identifying the person information, a person label to which the same person is given the same label, and a person feature. The person information is registered in the person table. The person ID is a number assigned to the information storage unit 450 after being counted up. The person ID may include characters and symbols. The person feature includes four color tables of 0 ° direction, 90 ° direction, 180 ° direction, and 270 ° direction. In FIG. 8, the table name of the color table is shown. Note that the person information of the person ID “06” is not assigned a person label, but is the person information to be processed, so that the association with the person information of the person IDs “01” to “05” has not been completed. . An appropriate person label is assigned to the person information of the person ID “06” by the association process described later. When the information registration unit 440 finishes the process of step S107, the video acquisition unit 410 starts the process of step S101. In this way, personal information representing the characteristics of a person is registered in the information storage unit 450 in order.

  Next, the color table update process described above will be described with reference to FIG.

FIG. 10 is a flowchart illustrating an example of the color table update process.
First, the color information generation unit 430 cuts out a person image as a determination target from the image (step S201). For example, the color information generation unit 430 separates the person image that is the area of the person M and the background from the image using background difference processing. Next, the color information generation unit 430 normalizes the person image (step S202). For example, the color information generation unit 430 changes the size of the person image so that the length from the top of the head to the shoe sole becomes a predetermined length.

  Next, the color information generation unit 430 identifies the direction in which the person is facing (step S203). For example, the color information generation unit 430 specifies the direction of the person using the moving direction and face direction of the person M in the human image, the ratio of head hair and skin, the positional relationship thereof, and the like. The color table in the direction closest to the direction of the identified person becomes the update target and color table. Next, the color information generation unit 430 extracts the color information of each region for each direction, and updates the color table (step S204). Specifically, the center position of the person image is adjusted to be arranged at a predetermined position (for example, the origin), color information is extracted in order from the upper left area, and is registered in the extracted color information color table. . Thereby, as shown in FIG. 5, the color information is registered in the corresponding color information column in the color table.

  Next, the operation of the same object detection device 400 will be described with reference to FIGS.

FIG. 11 is a flowchart illustrating an example of the association process executed by the same target detection device 400.
First, the information extraction unit 460 determines whether new person information has been registered (step S301). For example, the information extraction unit 460 periodically monitors the information storage unit 450 and determines whether or not personal information different from the previous time is registered. When the information extraction unit 460 determines that new person information has been registered (step S301: YES), the information extraction unit 460 extracts the registered person information as a comparison source (step S302). On the other hand, when the information extraction unit 460 determines that no new person information is registered (step S301: NO), the information extraction unit 460 returns to the process of step S301.

  After extracting the comparison source person information, the information extraction unit 460 extracts the person information to be compared with the person information as a comparison destination (step S303). The person information of the comparison destination may be all the person information stored in the information storage unit 450, or a person registered within a predetermined time, for example, by adding information such as time and travel route to the person information Information may be extracted as a comparison destination.

  Next, the difference determination unit 470 performs similarity calculation processing (step S304). Although details will be described later, the similarity calculation process is a process of calculating the similarity between any one of the comparison source person information and the comparison target person information. By the similarity calculation process, the similarity between any one of the comparison source person information and the comparison target person information is calculated. When the similarity calculation process is completed, next, the difference determination unit 470 determines whether or not the similarities with all the comparison destinations have been calculated (step S305). When the similarity determination unit 470 determines that the similarity with all the comparison destinations has not been calculated (step S305: NO), the similarity determination process is executed again. As a result, the similarity between the comparison source person information and all of the comparison target person information is calculated.

  When the similarity determination unit 470 determines that the similarity with all the comparison destinations has been calculated (step S305: YES), the difference determination unit 470 determines whether the maximum similarity is equal to or greater than a predetermined threshold (step S306). . When the difference determination unit 470 determines that the maximum similarity is equal to or greater than a predetermined threshold (step S306: YES), the information update unit 480 compares the same label with the label of the person information having the maximum similarity. The original person information is written (step S307). That is, it is determined that the person information of the comparison source represents the same person as the person information of the comparison destination having the maximum similarity. On the other hand, when the difference determination unit 470 determines that the maximum similarity is not greater than or equal to the predetermined threshold (step S306: NO), the information update unit 480 writes a new label in the comparison source personal information (step S308). ). That is, it is determined that the comparison source person information represents a person different from the comparison destination person information having the maximum similarity. When the information update unit 480 finishes the process of step S307 or S308, the information extraction unit 460 executes the process of step S301.

  Next, the similarity calculation process described above will be described with reference to FIGS.

FIG. 12 is a flowchart illustrating an example of similarity calculation processing. FIG. 13 is a diagram for explaining an example of similarity calculation processing.
First, as shown in FIG. 12, the difference determination unit 470 acquires a color table in the same direction (step S401). Specifically, the difference determination unit 470 selects one direction from the color tables for each direction included in each of the comparison-source person information and the comparison-destination person information, and selects the selected direction. Get the color table. As described with reference to FIG. 8, since the person information includes a plurality of color tables that differ depending on the orientation of the person, the difference determination unit 470 selects one of the color tables. Regarding the direction to be selected, a priority order is determined in advance for each direction, and a color table having a higher priority order is selected. The difference determination unit 470 may select all directions. In this case, the final similarity is calculated by obtaining and averaging the similarity for all orientations.

Next, the difference determination unit 470 calculates the similarity of the color information of each region for each region (step S402). As described with reference to FIG. 8, the color table has a plurality of time-series color information for one area. As a comparison method, for example, an average value of all color information To obtain the similarity of color information, to obtain the similarity of all combinations of all color information, and to adopt the highest similarity among them, the time series of multiple color information is similar There is a direction to check whether or not. Further, for example, there is a method of using the Euclidean distance between two points in the RGB space for calculating the similarity. The Euclidean distance dis between two points of the color information (R1, G1, B1) and the color information (R2, G2, B2) is expressed by the following formula (1).

Here, when the distance is obtained and the similarity degree cor is defined as in the following formula (2), for example, when it is similar, the value approaches “1” and is not similar A value that approaches the value “0” can be obtained.

Next, the difference determination unit 470 calculates the overall similarity using the weighting factor (step S403). Specifically, the difference determination unit 470 adds the weights corresponding to the respective regions to the similarities obtained in the respective regions, adds them, and averages them to obtain the similarity of the whole person image. For example, as illustrated in FIG. 13, when comparing the color table of the person ID “01” that is the comparison destination based on the imaging device 100 and the color table of the person ID “06” that is the comparison source based on the imaging device 200, The similarity cor_all of the whole person image can be obtained by the following formula (3). Note that i and j correspond to x and y of the region position, respectively. One of wa _{i, j} and wb _{i, j} corresponds to the weighting factor of the comparison source, and the other of wa _{i, j} , wb _{i, j} corresponds to the weighting factor of the comparison destination. For example, the similarity cor_all between the person ID “01” and the person ID “06” is the maximum compared to the remaining similarity cor_all, and the similarity cor_all between the person ID “01” and the person ID “06” is equal to or greater than a predetermined threshold. Then, “person A” is registered in the person label of the person ID “06” in FIG.

  As described above, according to the present embodiment, when a person image including the person M as a determination target is divided into a plurality of areas and the time-series color information of each area is compared, it is compared with other areas. Thus, in a region where the change is large, the weighting factor representing the reliability of the similarity is lowered to determine whether the determination targets are the same. Thereby, in the region where the change is large, the influence on the determination of the difference is small, and it is easy to determine that it is the same in the determination of the difference of the whole person image. Therefore, erroneous detection of the same object can be suppressed as compared with the case where the weighting coefficient representing the reliability of the similarity is not lowered for the region where the change is large. In other words, the detection accuracy of the same object can be improved.

  In particular, when a large region including the entire body of a person is set to cope with a large posture change, if color information is expressed by a histogram, the person is different because the colors of the upper body and the lower body are opposite. May be mistakenly detected as the same person. However, according to the present embodiment, such erroneous detection can be suppressed.

  The preferred embodiments of the present invention have been described in detail above, but the present invention is not limited to the specific embodiments according to the present invention, and various modifications are possible within the scope of the gist of the present invention described in the claims.・ Change is possible.

In addition, the following additional notes are disclosed regarding the above description.
(Supplementary Note 1) A target detection unit that detects a first determination target and a second determination target from each of a first captured image captured in time series and a second captured image captured in time series, and When the first determination target and the second determination target are each divided into a plurality of areas and the color information of the corresponding areas is compared with each other, the change of the color information in the time series of the areas to be compared Using a second weighting factor that is smaller than the first weighting factor used for any one of the regions in a region that is larger than any one of the regions. The same object detection apparatus which has a different judgment part which judges whether a judgment object is the same.
(Additional remark 2) It has the color information generation part which produces | generates the said color information for every moving direction with respect to each moving direction of the said 1st judgment object and the said 2nd judgment object, The said same judgment part is the same The same object detection device as set forth in appendix 1, wherein when comparing the color information in the moving directions of the first and second determinations, it is determined whether or not the first determination object and the second determination object are the same.
(Additional remark 3) The said difference determination part calculates the 1st similarity of the said color information for every area | region, The said 1st weight coefficient corresponding to each area | region or the said 1st similarity for every area | region or The second similarity is calculated by multiplying the second weighting factor and adding them together to calculate the second similarity between the first determination object and the second determination object, and the second similarity is the same. 3. The same object detection apparatus according to appendix 1 or 2, wherein the first determination object and the second determination object are determined to be the same when a determination threshold value is exceeded.
(Supplementary Note 4) A first determination object and a second determination object are detected from each of a first captured image captured in time series and a second captured image captured in time series, and the first determination is performed. When the target and the second determination target are each divided into a plurality of areas and the color information of the corresponding areas is compared with each other, the change of the color information in the time series of the areas to be compared is any Using the second weighting factor smaller than the first weighting factor used for any one of the regions, the first determination target and the second determination target are the same for the region larger than the region The same object detection method in which a computer executes a process of determining whether or not.
(Additional remark 5) It has the process which produces | generates the said color information for every moving direction with respect to each moving direction of the said 1st judgment object and the said 2nd judgment object, The said judgment process has the same movement direction 5. The same object detection method according to appendix 4, wherein when the color information is compared, it is determined whether or not the first determination object and the second determination object are the same.
(Additional remark 6) The said determination process calculates the 1st similarity of the said color information for every area | region, The said 1st weighting coefficient corresponding to each area | region or the said 1st similarity for every area | region By multiplying and adding the second weighting factor and averaging, the second similarity of the first determination object and the second determination object is calculated, and the second similarity is determined to be the same 6. The method of detecting the same object according to appendix 4 or 5, wherein the first determination object and the second determination object are determined to be the same when the threshold value is equal to or greater than a threshold value.
(Supplementary note 7) A first determination object and a second determination object are detected from each of a first captured image captured in time series and a second captured image captured in time series, and the first determination is performed. When the target and the second determination target are each divided into a plurality of areas and the color information of the corresponding areas is compared with each other, the change of the color information in the time series of the areas to be compared is any Using the second weighting factor smaller than the first weighting factor used for any one of the regions, the first determination target and the second determination target are the same for the region larger than the region The same object detection program for causing a computer to execute processing for determining whether or not

S Same target detection system 100, 200, 300 Imaging device 400 Same target detection device 420 Target detection unit 430 Color information generation unit 470 Difference determination unit

Claims (5)

An object detection unit for detecting the first determination object and the second determination object from each of the first captured image captured in time series and the second captured image captured in time series;
When the first determination object and the second determination object are each divided into a plurality of areas and the color information of the corresponding areas is compared with each other, the color information in the time series of the areas to be compared is compared. Using the second weighting factor that is smaller than the first weighting factor used for any one of the regions in the region where the change is larger than any one of the regions, the first determination target and the second A judgment unit for judging whether or not the judgment targets are the same,
The same object detection device having A color information generation unit that generates the color information for each movement direction with respect to the movement direction of each of the first determination object and the second determination object;
The difference determination unit, when comparing the color information in the same movement direction, determines whether the first determination object and the second determination object are the same. The same object detection device according to 1.   The difference determination unit calculates a first similarity between the color information for each region, and the first weight coefficient corresponding to each region or the second weight for the first similarity for each region. By multiplying and adding the weighting factor and averaging, the second similarity between the first determination target and the second determination target is calculated, and the threshold is equal to or higher than the threshold for determining that the second similarity is the same. 3, the same object detection apparatus according to claim 1, wherein the first determination object and the second determination object are determined to be the same. A first determination object and a second determination object are detected from each of the first captured image captured in time series and the second captured image captured in time series;
When the first determination object and the second determination object are each divided into a plurality of areas and the color information of the corresponding areas is compared with each other, the color information in the time series of the areas to be compared is compared. Using the second weighting factor that is smaller than the first weighting factor used for any one of the regions in the region where the change is larger than any one of the regions, the first determination target and the second To determine whether the subject of judgment is the same,
Same object detection method in which processing is executed by computer. A first determination object and a second determination object are detected from each of the first captured image captured in time series and the second captured image captured in time series;
When the first determination object and the second determination object are each divided into a plurality of areas and the color information of the corresponding areas is compared with each other, the color information in the time series of the areas to be compared is compared. Using the second weighting factor that is smaller than the first weighting factor used for any one of the regions in the region where the change is larger than any one of the regions, the first determination target and the second To determine whether the subject of judgment is the same,
The same object detection program for causing a computer to execute processing.

Images