Image recognition support device and program

Abstract

To specify a region in which an event has occurred based on a motion pattern of a group and support improvement of the accuracy of recognizing target objects in the region.SOLUTION: An image recognition support device according to the present disclosure acquires a recognition result obtained by subjecting an input image received from an imaging device including an image quality adjustment unit that adjusts an image quality using an image quality adjustment parameter to processing of recognizing a plurality of target objects by an image recognition device, acquires, based on the recognition result, positional information about the plurality of target objects in the input image, calculates a speed vector of each of the plurality of target objects, sets an extension of each of the speed vectors based on the magnitude of speed vector, specifies a priority area based on an intersection of a plurality of set extensions, determines a priority order of the target objects based on the specified priority area, determines the image quality adjustment parameter for the input image to be subjected to the recognition processing such that a rate of recognition of a target object having a higher priority order in the image recognition device is improved, and sets the determined image quality adjustment parameter in the image quality adjustment unit.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present disclosure relates to an image recognition assistance device and a program. [Background technology]

[0002] Patent Document 1 discloses a state recognition device that selects in advance the measurement range and measurement targets for state recognition processing based on interaction information between measurement targets extracted from flow information on the general movements of the measurement targets. This state recognition device acquires people flow information by analyzing image data captured by a camera, and outputs areas where there are changes in people flow as candidate recognition areas.

[0003] The state recognition device calculates the degree of people flow interaction for each recognition area candidate based on the amount of change in the person's movement line relative to the recognition area candidate calculated from the person's movement line information and the degree of attention to the recognition area candidate calculated from the person's line of sight information.The recognition area candidate with the high degree of people flow interaction is determined as the target for executing the state recognition process.

[0004] The state recognition device also extracts the degrees of interaction between people to determine the person for whom state recognition processing is to be performed. The state recognition device calculates the degrees of interaction between the person in the candidate recognition area and the surrounding people, with a person who is in the candidate recognition area immediately after a change in people flow is detected as the center position, and determines the person whose final sum of the degrees of interaction is higher than a threshold as the person for whom state recognition processing is to be performed. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Publication No. 2020-135099 Summary of the Invention [Problem to be solved by the invention]

[0006] In Patent Document 1, the direction of the flow line is used as a method for calculating changes in the flow line. The direction of each person's flow line is recorded at every time, and the degree of change in the direction of the flow line near the candidate recognition area is taken as the amount of change in the flow line. For example, a value range is set so that if a person who was moving toward the back continues to move toward the back, the amount of change is 0, and if the movement direction changes in the exact opposite direction toward the front, the amount of change is at its maximum. The sum of the degree of change in the movement direction of each person near the candidate recognition area, etc., is determined as the degree of interaction of the people flow with respect to the candidate recognition area.

[0007] When a notable event occurs among a group of people, the people may behave in a similar manner due to group psychology. It is necessary to identify the area where the event is occurring from the movement patterns of such a group and detect the object within that area with high recognition accuracy. In Patent Document 1, the target for executing state recognition processing is selected based on the degree of change in the direction of each person's movement line, not based on the movement patterns of the group.

[0008] In view of the above-mentioned problems, an object of the present disclosure is to provide an image recognition support device and a program that can identify an area where an event is occurring based on the movement patterns of a group and support improving the recognition accuracy of objects within that area. [Means for solving the problem]

[0009] The image recognition assistance device of the present disclosure includes: a recognition result acquisition unit that acquires recognition results obtained by performing recognition processing on multiple objects using an image recognition device on an input image input from an imaging device equipped with an image quality adjustment unit that adjusts image quality using image quality adjustment parameters; a position information acquisition unit that acquires position information of the multiple objects on the input image based on the recognition results; an extension line setting unit that calculates velocity vectors of the multiple objects and sets extension lines of each of the velocity vectors based on the magnitudes of the velocity vectors; a priority area identification unit that identifies a priority area based on the intersection of the multiple extension lines that have been set; a priority determination unit that determines the priority of the objects based on the identified priority area; and a parameter setting unit that determines image quality adjustment parameters for the input image on which recognition processing is performed so as to improve the recognition rate, which indicates the likelihood of recognition results for objects with high priority in the image recognition device, and sets the image quality adjustment parameters determined in the image quality adjustment unit.

[0010] The program of the present disclosure causes a computer to perform the following processes: a process of acquiring recognition results of an input image input from an imaging device equipped with an image quality adjustment unit that adjusts image quality using image quality adjustment parameters, where the input image is subjected to recognition processing of multiple objects by an image recognition device; a process of acquiring position information of the multiple objects on the input image based on the recognition results; a process of calculating velocity vectors of the multiple objects and setting extension lines of each of the velocity vectors based on the magnitude of the velocity vectors; a process of identifying a priority area based on the intersection of the multiple extension lines that have been set; a process of determining a priority order for the objects based on the identified priority area; and a process of determining image quality adjustment parameters for the input image on which the recognition processing is performed so as to improve the recognition rate in the image recognition device, which indicates the likelihood of recognition results for objects with high priority, and setting the determined image quality adjustment parameters to the image quality adjustment unit. [Effects of the Invention]

[0011] The present disclosure makes it possible to identify an area where an event is occurring based on the movement patterns of a group, and to help improve the accuracy of recognizing objects within that area. [Brief explanation of the drawings]

[0012] [Figure 1] 1 is a block diagram showing the overall configuration of an image recognition system including an image recognition support device according to an embodiment. [Figure 2] 2 is a block diagram showing an example of the configuration of the imaging device of FIG. 1. FIG. [Figure 3] FIG. 2 is a block diagram showing an example of the configuration of the image recognition support device of FIG. 1. [Figure 4] FIG. 10 is a diagram illustrating an example of a behavior pattern of an object. [Figure 5] FIG. 10 is a diagram showing another example of the behavior pattern of the object. [Figure 6] FIG. 5 is a diagram showing extension lines of velocity vectors in the example of FIG. 4. [Figure 7] FIG. 10 is a diagram showing the positional relationship between a target object and surrounding people. [Figure 8] FIG. 8 is a diagram showing extensions of the velocity vectors of the people T1 to T7 in FIG. [Figure 9] FIG. 10 is a diagram illustrating grouping of intersections of extensions of the velocity vectors of each person. [Figure 10] FIG. 10 is a diagram showing priority regions. [Figure 11] FIG. 10 is a flowchart showing the flow of a process for determining a priority order in the image recognition support process according to the embodiment. [Figure 12] 10A and 10B are diagrams illustrating an example of a change in recognition rate in accordance with a change in an image quality adjustment parameter. [Figure 13] FIG. 13 is a diagram illustrating a process of determining a priority order taking into consideration the line of sight of an object. [Figure 14] FIG. 10 is a flowchart showing the flow of processing for setting image quality adjustment parameters. DETAILED DESCRIPTION OF THE INVENTION

[0013] Hereinafter, specific embodiments of the present disclosure will be described in detail with reference to the drawings. In each drawing, the same elements are denoted by the same reference numerals, and for clarity of explanation, duplicated explanations will be omitted as necessary.

[0014] The present invention relates to an image recognition support device that supports the improvement of recognition accuracy in an image recognition system including an imaging device such as a surveillance camera and a recognition device that performs object recognition processing on images captured by the imaging device. Many general image recognition systems have the function of capturing an image of an unspecified number of objects and recognizing multiple objects. When multiple objects exist in a captured image, the priority of detection may change depending on the behavioral patterns of the objects.

[0015] For example, when a notable event occurs while a group of people are gathered, people often behave in a group, either congregating at the location where the event occurred or moving away, due to herd psychology. In an embodiment, the area where the event is occurring is identified from the behavioral patterns of such a group, and people in the identified area are given a high priority as important people who should be noted. In an embodiment, image processing parameters suitable for image recognition processing of important people with high priority are set to improve the accuracy (recognition rate) of image recognition.

[0016] An example of a situation where priorities change based on the behavioral patterns of multiple objects is a surveillance camera system installed in busy shopping districts or tourist spots where many people gather, with the aim of monitoring an unspecified number of people. Such a surveillance camera system is required not only to monitor the status of people within the entire coverage area, but also to monitor in detail special situations such as crimes, accidents, and television filming. For example, if there is a dangerous person carrying a knife, people around the dangerous person will often move away from the dangerous person in an attempt to escape. In this case, a dangerous person near the starting point of the surrounding people's movements can be determined to have a high priority as a target of attention.

[0017] Furthermore, when a celebrity is present during a television shoot, people around the celebrity often move in a direction that approaches the celebrity. In this case, by increasing the priority of people who are near the end point of the surrounding people's movements, it is possible to monitor any unusual events without missing them. The embodiment has a feature of determining priorities by focusing on the behavioral patterns of multiple objects and adjusting image quality to favor the recognition process of objects with high priority.

[0018] Fig. 1 is a block diagram showing the overall configuration of an image recognition system 100 including an image recognition support device 10 according to an embodiment. As shown in Fig. 1, the image recognition system 100 includes the image recognition support device 10, an imaging device 20, an image recognition device 30, a display device 40, and an input device 50. The image recognition system 100 has a function of displaying an image captured by the imaging device 20 for visual confirmation by a user. The image recognition system 100 also has a function of recognizing an object included in the captured image and displaying it so that it can be visually recognized by the user.

[0019] An example will be described below in which the image recognition system 100 configures a surveillance camera system that monitors an unspecified number of people in a busy shopping district where many people are present. In this example, multiple pedestrians passing through the shopping district are the objects to be recognized by the image recognition device 30. Note that, although the objects in this example are people, they are not limited to this and may also be vehicles such as cars and motorcycles operated by people.

[0020] <Imaging device 20> The imaging device 20 captures an image of a scene including multiple objects, and generates display image data and recognition image data from the captured image data. Fig. 2 is a block diagram showing an example configuration of the imaging device 20 of Fig. 1. As shown in Fig. 2, the imaging device 20 includes a camera unit 21, a signal processing unit 22, an image output unit 23, and a recognition signal processing unit 24.

[0021] Camera unit 21 includes, as an imaging element, for example, a CCD (Charge Coupled Device) image sensor or a CMOS (Complementary Metal Oxide Semiconductor) sensor. Camera unit 21 may also include a lens group (imaging optical system) including a zoom lens and a focus lens, an iris diaphragm, a mechanical shutter, etc. Camera unit 21 adjusts the level of an image signal based on an image formed by the imaging optical system according to a predetermined amplification gain, performs A / D conversion, and sequentially outputs the signal to signal processing unit 22 as shooting data.

[0022] The signal processing unit 22 acquires image data continuously captured by the camera unit 21 and performs various image processing on image data in units of frames (hereinafter referred to as image frames). The signal processing unit 22 can generate image frames in accordance with a predetermined signal format for display. For example, when generating image data in full high-definition format, the signal processing unit 22 performs white balance processing, gamma processing, etc. on each image frame, and then performs enlargement / reduction processing to an image size of 1920 x 1080 pixels. The signal processing unit 22 passes the image data that has been subjected to each processing to the image output unit 23.

[0023] The image output unit 23 can acquire the recognition results from the image recognition device 30 (described later) via the image recognition support device 10. The image output unit 23 can process the image data using the recognition results to generate display image data. The image output unit 23 outputs the generated display image data to the display device 40. The display device 40 is a liquid crystal display device or the like with an image display function. The display device 40 can display a highly visible captured image and a display image including the recognition result based on the display image data.

[0024] The image output unit 23 may use the recognition result to generate information to be displayed superimposed on the captured image. The display device 40 can superimpose, on the captured image, for example, a frame surrounding the area of ​​the recognized object, text information corresponding to the type of the recognized object, recognition determination information such as a recognition rate, and the like.

[0025] The image output unit 23 may also generate auxiliary display information such as a menu setting display, which is used when the user inputs setting information using the input device 50 described below. The image output unit 23 can superimpose the menu setting display on the captured image.

[0026] The image data processed by the signal processing unit 22 is used for display on the display device 40 and is also supplied to the recognition signal processing unit 24. The recognition signal processing unit 24 converts the image data processed by the signal processing unit 22 into recognition image data suitable for object recognition processing. The recognition image data is supplied to the image recognition device 30 and used to recognize the object. Note that the image data generated by the camera unit 21 may be supplied directly to the recognition signal processing unit 24 without going through the signal processing unit 22. Furthermore, a signal in the middle of being processed by the signal processing unit 22 may be supplied to the recognition signal processing unit 24.

[0027] The recognition signal processing unit 24 converts image data into image data for recognition by performing image quality adjustment processing based on the image quality adjustment parameters. In other words, the recognition signal processing unit 24 can also be considered an image quality adjustment unit that adjusts image quality using the image quality adjustment parameters. The image recognition support device 10, which will be described later, generates image quality adjustment parameters that affect the recognition rate of the object and are used when the recognition signal processing unit 24 performs image quality adjustment processing. Such image quality adjustment parameters include a luminance gain and tone mapping characteristics for controlling the luminance (brightness) of the captured image, an aperture gain for edge enhancement, and the like.

[0028] <Image recognition device 30> The image recognition device 30 performs image recognition processing on the recognition images supplied from the recognition signal processing unit 24 and outputs the recognition results to the image recognition assistance device 10. Recognition images are continuously input to the image recognition device 30, and recognition processing is continuously performed at any time. The recognition results include the presence or absence of an object, the type of object, the area or position of the object, and the recognition rate. The presence or absence of an object is information indicating whether an object has been recognized, i.e., identified, by the image recognition processing on the recognition image. The type of object is information indicating the type of the recognized object. The area of ​​the object is a set of coordinates that defines the range of the area including the recognized object in the recognition image. The area of ​​the object is, for example, a range specified by pixel values ​​in an XY coordinate system. The position of the object is, for example, a representative point such as the center coordinates of the object recognized in the recognition image.

[0029] The recognition rate is an example of the degree of accuracy of recognition by image recognition. In other words, the recognition rate is numerical information indicating the presence or absence, type, and recognition accuracy of the area of ​​an object recognized by image recognition processing in an image for recognition. The recognition rate may be expressed, for example, as a range from 0 to 100%. Furthermore, the recognition rate may be calculated using, for example, a threshold indicating the similarity with the object, the number of stages passed by a classifier, etc. When multiple recognition objects are recognized, the recognition result may generate a set of type, area, and recognition rate for each object.

[0030] The image recognition device 30 is hardware or software, or a combination thereof, capable of executing a known image recognition process. For example, the image recognition device 30 is realized by executing a known image recognition processing program on a computer. The image recognition device 30 may be redundantly configured on multiple computers, and each functional block may be realized by multiple computers. The image recognition device 30 may also be realized as a client-server system, a cloud computing system, or the like, in which each system is connected via a communication network. The functions of the image recognition device 30 may also be provided in a SaaS (Software as a Service) format. Alternatively, the image recognition device 30 may be realized on the same computer as the image recognition assistance device 10.

[0031] The image recognition process by the image recognition device 30 may involve storing multiple images for each object and recognizing the object using pattern matching. In this case, deep learning may be performed using multiple images captured from various angles to create a model for object recognition. It is generally known that the recognition rate of such image recognition processes varies depending on image characteristics, such as the ease of distinguishing between the object to be recognized and the background in the recognition image.

[0032] The image recognition device 30 also has a function of tracking each recognized object using known techniques such as motion compensation between image frames. The image recognition device 30 can link identification information, such as an identification number for identifying the same object, with location information and provide the linked information to the image recognition assistance device 10. The above-mentioned recognition result may also include the identification information.

[0033] The image recognition device 30 may recognize the direction of the face or body of a person from an image of the detected person. The image recognition device 30 may also detect the gaze direction from the positional relationship between the characteristic parts of the person's eyes and the outer edge of the iris. The image recognition device 30 may estimate the gaze direction of the person by recognizing the direction in which the person's face or body is facing (forward, up, down, sideways, diagonally upward, diagonally downward, etc.). Note that the method for detecting the face direction or gaze may be any known method and is not particularly limited. The above-mentioned recognition result may also include information on the direction or gaze direction of the detected person. Hereinafter, information on the direction of the person and information on the gaze direction will be referred to as "information on the gaze."

[0034] <Image Recognition Support Device 10> Image recognition support device 10 determines and sets image quality adjustment parameters in accordance with the image recognition results for the recognition image output from image recognition device 30. Recognition signal processing unit 24 further adjusts the image quality of the recognition image using the set image quality adjustment parameters, and supplies the adjusted recognition image to image recognition device 30. When image recognition support device 10 obtains the recognition results for the adjusted recognition image, it again determines and sets image quality adjustment parameters.

[0035] In this way, image recognition support device 10 performs feedback control of image quality adjustment parameters according to the recognition results for the recognition image. Image recognition support device 10 repeats the feedback control until the recognition rate of the object in the recognition image by image recognition support device 10 stabilizes at a high level. This makes it possible to increase the recognition rate of the object in the recognition image.

[0036] Here, the multiple objects contained in the captured image differ in shooting conditions such as brightness and background. Therefore, the image quality adjustment parameters that optimize the recognition rate often differ for each object. Therefore, it is necessary to prioritize the multiple objects and perform feedback control of the image quality adjustment parameters so that the recognition rate of the object with the highest priority is high.

[0037] In the embodiment, priorities are determined based on the behavioral patterns of multiple objects, and image quality adjustment is performed to favor recognition processing of objects with high priorities. For example, if a behavioral pattern of multiple people leaving at the same time is observed, the object for which feedback control of image quality adjustment is performed is set to a person who is at a location that is the starting point, making it possible to obtain recognition results that are in line with the purpose of the image recognition device.

[0038] FIG. 3 is a block diagram showing an example of the configuration of the image recognition support device 10 of FIG. 1. Here, an example in which the image recognition support device 10 is implemented by a single computer is shown, but this is not limiting. The image recognition support device 10 may be implemented redundantly by multiple computers, and each functional block may be implemented by multiple computers. Alternatively, all or part of the functions of the image recognition support device 10 may be implemented by a general-purpose or dedicated circuit such as a semiconductor device. In these cases, the image recognition support device 10 may be communicably connected to the image capture device 20 and the image recognition device 30 via a communication network. Note that any combination of the image capture device 20, the image recognition device 30, and the image recognition support device 10 may be implemented by a single device, or all of these may be implemented by a single device.

[0039] The image recognition support device 10 includes a processing unit 1, a memory unit 2, and an IF (Interface) unit 3. The memory unit 2 includes a non-volatile memory device such as a hard disk or flash memory, and a memory such as RAM (Random Access Memory), i.e., a volatile memory device. The memory unit 2 stores an image recognition support program, recognition results, and image quality adjustment history. The image recognition support program is a computer program that implements the processing of an image recognition support method according to an embodiment. The recognition results include the recognition rate and position information for each object, identification information for identifying the same object, etc. The image quality adjustment history is a history of image quality adjustment parameters determined and set by the image recognition support device 10. The IF unit 3 is an interface circuit that transmits and receives data between the image recognition support device 10 and the outside.

[0040] The processing unit 1 is a processing device that executes each process of the image recognition support device 10. The processing unit 1 is, for example, a processor such as a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), an FPGA (Field-Programmable Gate Array), or a quantum processor (quantum computer control chip). The processing unit 1 loads an image recognition support program from the storage unit 2 into memory and executes it. As a result, the processing unit 1 realizes the functions of a recognition result acquisition unit 11, a position information acquisition unit 12, an extension line setting unit 13, a priority area identification unit 14, a priority order determination unit 15, a parameter setting unit 16, and an attention area estimation unit 17, and generates the image quality adjustment parameters described above. Some or all of the components of the processing unit 1 may be realized, for example, by a general-purpose or dedicated circuit realized in a semiconductor device.

[0041] The processing unit 1 can determine the object with the highest priority according to the following two examples. (Example 1) High-priority attention targets are predicted and determined based on the behavioral patterns of nearby people, such as the direction and speed of movement (velocity vector) of nearby objects and the direction of their faces (gaze). (Second example) The intersection points of the extensions of the velocity vectors of multiple objects in the vicinity are grouped to determine a priority area, and then the object of interest with the highest priority is determined within the priority area.

[0042] The recognition result acquisition unit 11 acquires the recognition results obtained by performing image recognition processing of multiple objects on the image for recognition by the image recognition device 30. The position information acquisition unit 12 acquires position information of the multiple objects on the image for recognition based on the acquired recognition results. As described above, the position information of the objects can be, for example, the center coordinates of the objects recognized in the image for recognition.

[0043] The extension line setting unit 13 calculates the velocity vectors of the multiple objects, and sets extension lines for each velocity vector based on the magnitude of the velocity vector. Specifically, the extension line setting unit 13 first calculates the movement direction of each object from the transition of the positions (coordinates) of each object between consecutive image frames. The extension line setting unit 13 can calculate the movement direction of each object using, for example, the identification information of each object input from the image recognition device 30 and the position information associated therewith.

[0044] Furthermore, the extension line setting unit 13 can calculate the speed of movement of each object from the transition of the position (coordinates) of each of the multiple objects between consecutive image frames. That is, the extension line setting unit 13 can obtain the velocity vector of each object. The velocity vector is a vector that represents the direction and magnitude of change in the position of the object in the recognition image over time.

[0045] The extension line setting unit 13 may set an extension line based on the velocity vector of an object whose magnitude is equal to or greater than a predetermined value among the velocity vectors of multiple objects. Here, "equal to or greater than the predetermined value" refers to, for example, the moving speed of the object indicated by the magnitude of the velocity vector being equal to or greater than 6 km / h. The extension line includes an extension in the moving direction of the object based on the velocity vector and an extension in the opposite direction to the moving direction.

[0046] Fig. 4 is a diagram showing an example of a behavior pattern of an object. Fig. 4 shows a recognition image P0 captured by an imaging device 20 monitoring pedestrians on sidewalks along both sides of a roadway, for example, in a busy shopping district. In Fig. 4, it is assumed that a "dangerous person" is present as the target of attention at the far end of the crosswalk. The target of attention is indicated by a black circle.

[0047] In Fig. 4, it is assumed that eight people P1 to P8 are included in the recognition image P0. In the example of Fig. 4, people P1 to P5 are on the same side of the sidewalk as the target of interest, and people P6 to P8 are on the opposite side of the sidewalk from the target of interest. Person P6 faces the target of interest across the sidewalk. The positions of people P1 to P8 in the first image frame are indicated by dashed circles. It is assumed that in the second image frame, which is generated a predetermined time interval after the first image frame, people P1 to P8 have moved to the positions indicated by solid circles.

[0048] Because the target of interest is a dangerous person, all of the people P1 to P5 on the sidewalk on the same side as the target of interest are moving in a direction away from the target of interest. In addition, person P6, who is opposite the target of interest across the sidewalk, is also moving in a direction away from the target of interest. The movement direction of each of the people P1 to P6 is indicated by the direction of the dashed arrow in the recognition image P0. The size of the dashed arrow indicates the movement speed of each of the people P1 to P6.

[0049] Note that people P7 and P8, who are on the sidewalk opposite the target of interest, barely move between the first and second image frames, and their positions remain almost unchanged. For this reason, the dashed circle and solid circle representing people P7 and P8 almost overlap. The movement speed of people P7 and P8 is zero.

[0050] If the image recognition device 30 has a function of recognizing the facial direction or line of sight of a person, the attention area estimation unit 17 can estimate an area that the detected person is looking at, based on information about the line of sight input from the image recognition device 30. For example, as shown in Fig. 4, the attention area estimation unit 17 can estimate that the lines of sight of persons P7 and P8 are directed toward a target of attention, based on the facial directions of persons P7 and P8 who are not moving.

[0051] Fig. 5 is a diagram showing another example of a behavior pattern of a target object. Fig. 5 shows a recognition image P0 captured by the same imaging device 20 as Fig. 4. In Fig. 5, it is assumed that a "famous person" such as a celebrity is present as the target of attention at the far end of the crosswalk. The target of attention is indicated by a black circle.

[0052] As in FIG. 4, FIG. 5 includes eight people P1 to P8. People P1 to P5, who are on the same side of the sidewalk as the target of interest, are all moving in a direction approaching the target of interest. Furthermore, person P6, who is across the sidewalk from the target of interest, is also moving in a direction approaching the target of interest. The movement direction of each person P1 to P6 is indicated by the direction of the dashed arrow in the recognition image P0. The size of the dashed arrow indicates the movement speed of each person P1 to P3. Because people P1 to P4 and P6 recognize the target of interest, as shown in FIG. 5, the movement speed of people P1 to P4 and P6 is greater than the movement speed of person P5. Furthermore, people P7 and P8, who are on the opposite side of the sidewalk from the target of interest, are barely moving, and it is estimated that their gaze is directed toward the target of interest.

[0053] As a first example, the priority determination unit 15 can predict and determine a target of interest with a high priority based on the behavioral patterns of people around the target of interest, such as the direction and speed of movement of people and the orientation of their faces. The priority determination unit 15 can determine a target of interest with a high priority using existing technology, such as a prediction model using supervised machine learning. For example, the priority determination unit 15 can use a prediction model trained using supervised learning data including the velocity vectors of people around the target of interest and position information relative to the target of interest.

[0054] The input device 50 includes a keyboard, a mouse, a touch panel, etc., and allows a user to input setting information. The setting information is, for example, information for setting a "predetermined range" used when grouping intersections, as described below, or a "threshold value" used in determining the speed of an object. The setting information may also be information for setting the range of nearby people whose behavior patterns are to be detected when determining a high-priority target of interest in the first example above. The image recognition assistance device 10 can receive setting information from the input device 50 via the IF unit 3. Note that if these values ​​do not need to be changed, they may be set in advance.

[0055] When the priority determination unit 15 cannot determine the target of interest using a prediction model, the second example is used. In the second example, an algorithm is used to detect divergent or convergent behavior patterns of people around the target of interest.

[0056] As described above, the extension line setting unit 13 obtains the velocity vector of each object from changes in position coordinates between image frames. The velocity vector indicates the direction and speed of movement. In the second example, the extension line setting unit 13 sets an extension line for each of the calculated velocity vectors of the multiple objects. Specifically, the extension line setting unit 13 sets extension lines by extending both ends of each velocity vector.

[0057] Figure 6 illustrates the extension lines of the velocity vectors in the example of Figure 4. In Figure 6, the extension lines of the velocity vectors of persons P1 to P4 are extended only toward the target of interest. An area including multiple intersections of the extension lines is identified as a priority area AA, and objects present within the priority area AA are given a higher priority than objects outside the priority area.

[0058] The process of determining the target of interest in the second example will be described in detail below with reference to Figures 7 to 10. Figure 7 is a diagram showing the positional relationship between the target of interest and surrounding people. It is assumed that seven people, T1 to T7, are included in Figure 7. For the sake of explanation, it is assumed that person T1 is the target of interest. The target of interest is indicated by a black circle. The positions of each of people T1 to T7 in the first image frame are indicated by dashed circles. It is assumed that in the second image frame, which is obtained a predetermined time interval after the first image frame, each of people T1 to T7 has moved to the positions indicated by the solid circles.

[0059] The priority area identification unit 14 identifies a priority area based on the intersections of the multiple extension lines that have been set. FIG. 8 is a diagram showing extension lines obtained by extending the velocity vectors of the people T1 to T7 in FIG. 7. In FIG. 8, each extension line is shown as a two-dot chain line. The priority area identification unit 14 extracts intersections of the extension lines. Each intersection is shown as a hatched circle. This intersection is an intersection of diverging directions or an intersection of converging directions. A "diverging direction intersection" is an intersection where objects on each of the intersecting extension lines move away from the intersection. A "converging direction intersection" is an intersection where objects on each of the intersecting extension lines move toward the intersection. In FIG. 8, the convergent direction intersection is indicated by Ic, and the diverging direction intersection is indicated by Id.

[0060] When one of the multiple intersections is set as a central intersection, the priority area identification unit 14 extracts and groups all intersections that are within a predetermined range centered on the central intersection and that include at least one intersection other than the central intersection, and identifies a predetermined area that includes all the grouped intersections as a priority area AA. Figure 9 is a diagram illustrating grouping of intersections of extension lines of each person's velocity vector.

[0061] The predetermined range can be set arbitrarily depending on the detection purpose and specifications. For example, if the main purpose is to detect a dangerous person behaving suspiciously in a busy shopping district, the distance at which people in the vicinity can recognize the dangerous person's actions and behavior is thought to be about 10 meters. Information about the actions of people who are more than 10 meters away from the dangerous person and cannot recognize the dangerous person becomes noise in the detection of the dangerous person.

[0062] If we assume that the intersection of the extension lines of the velocity vectors is the starting point of the divergent behavior of each person who has identified a dangerous person, other intersections within a 10-meter radius from the intersection point often indicate divergent behavior by the same dangerous person. Therefore, the predetermined range can be set to a 10-meter radius from the intersection point. For example, the priority area identification unit 14 can calculate the size (number of pixels, etc.) on the image corresponding to 10 meters from the angle of view information of the imaging device 20, and use the calculated size on the image to group the intersections of the extension lines of the velocity vectors of each person T1 to T7.

[0063] As shown in Fig. 9, the predetermined range centered on the intersection point Id of the five diverging directions in Fig. 8 includes other intersection points. However, the predetermined range centered on the intersection point Ic of one converging direction in Fig. 8 does not include other intersection points. Therefore, in the example of Fig. 9, the intersection points Id of the five diverging directions are grouped. The intersection point Ic of one converging direction is not included in the grouping.

[0064] Specifically, if all of the grouped intersections are intersections of converging directions or intersections of diverging directions, the priority area identification unit 14 identifies the area as a priority area AA. In other words, depending on whether the movement direction of each person on the extension line is a direction away from or a direction toward the intersection, it can be determined whether the behavior pattern of the group is diverging behavior or converging behavior.

[0065] Fig. 10 is a diagram showing a priority area AA. In the example shown in Fig. 10, a rectangular area including five intersections Id of divergent directions becomes the priority area AA. Note that the shape of the area including all of the grouped intersections is not limited to a rectangle, and any shape may be adopted. Furthermore, since the distance at which the behavior or speech of the target of interest can be recognized varies depending on the environmental conditions such as the level of crowding and the time of day (daytime, nighttime, etc.), the predetermined range may be variable depending on the environmental conditions.

[0066] The priority determination unit 15 determines the priority of the objects based on the identified priority area AA. The priority determination unit 15 can assign a higher priority to the object (person T1) within the priority area AA than to the objects (persons T2 to T7) outside the priority area. If multiple objects exist within the priority area, the priority determination unit 15 can determine the priority of the objects based on the distance from the center point of the priority area AA. For example, the priority determination unit 15 may assign higher priority to objects closer to the center of the priority area AA.

[0067] It should be noted that people around a person of interest may be influenced by herd psychology, wanting to move faster. For example, a person who recognizes a dangerous person may run away from the dangerous person along with other people who also recognize the dangerous person. For this reason, the extension line setting unit 13 takes into consideration the moving speed of the object (the magnitude of the velocity vector) and sets an extension line of each velocity vector when the moving speed of the object is equal to or greater than a predetermined value. For example, the extension line setting unit 13 may calculate extension lines at both ends of the velocity vector when the moving speed of each object is equal to or greater than a reference value. The extension line setting unit 13 may not set an extension line of the velocity vector when the moving speed of each object is smaller than the reference value.

[0068] The speed at which a person normally walks (hereinafter referred to as "normal movement") is usually 1 km / h or more and less than 6 km / h. If the speed is 6 km / h or more, it is considered that the person is moving fast (hereinafter referred to as "high-speed movement"). Therefore, the reference value can be set to, for example, 6 km / h.

[0069] Furthermore, the people recognized in the recognition image may include people who are unaware of the person of interest, or people who notice the person of interest but do not pay attention to them. For this reason, the extension line setting unit 13 may calculate an extension line of the velocity vector when the movement speed of a predetermined percentage of the recognized objects is equal to or greater than a reference value. For example, the extension line setting unit 13 can calculate an extension line of the velocity vector when the movement speed of all the people recognized in the recognition image is equal to or greater than a reference value, or when the movement speed of 50% or more of the people recognized in the recognition image is equal to or greater than a reference value.

[0070] Here, a method for determining the priority will be described with reference to Fig. 11. Fig. 11 is a flow chart showing the flow of processing for determining the priority in the image recognition support processing according to the embodiment. Although not shown in Fig. 11, the recognition signal processing unit 24 generates image data for recognition using initial setting image quality adjustment parameters and outputs it to the image recognition device 30. The image recognition device 30 performs recognition processing on the input image data for recognition and generates a recognition result.

[0071] First, the image recognition support device 10 acquires the recognition results obtained by performing image recognition processing of multiple objects on the recognition image by the image recognition device 30. The image recognition support device 10 acquires position information of the multiple objects on the recognition image based on the acquired recognition results. The recognition rate and image quality adjustment history for each object are managed in the storage unit 2. The image recognition support device 10 calculates the movement direction and speed (velocity vector) from changes in the position information (step S101).

[0072] Then, it is determined whether the speed of each object is equal to or greater than a reference value (step S102). If the speed of each object is smaller than the reference value (step S102, NO), the image recognition support device 10 determines that "the object has no priority" (step S108, and ends the process). If the speed of each object is equal to or greater than the reference value (step S102, YES), an extension line of the velocity vector of each object is set (step S103).

[0073] Then, the image recognition support device 10 extracts intersections of the extension lines of the velocity vectors of each object (step S104). As described above, when one of the intersections is set as the central intersection, the image recognition support device 10 extracts all intersections that are within a predetermined range centered on the central intersection and that include at least one other intersection than the central intersection. Each intersection is a diverging intersection or a converging intersection.

[0074] Then, the image recognition support device 10 groups the extracted intersections (step S105). Next, it is determined whether all the intersections in the group are divergent intersections (step S106). If the result in step S106 is NO, it is determined whether all the intersections in the group are convergent intersections (step S107). If the result in step S107 is NO, the objects are not assigned a priority, and the process ends (step S108).

[0075] If the answer is YES in step S106, it can be determined that the behavior pattern of a group consisting of multiple objects is divergent behavior. Also, if the answer is YES in step S107, it can be determined that the behavior pattern of a group consisting of multiple objects is convergent behavior. In either case, the area containing all the grouped intersections is determined as the priority area AA. The image recognition assistance device 10 identifies a "target of interest" with a high priority (step S109). Specifically, objects included in the priority area are given a higher priority than objects outside the priority area. In this way, the priority of the objects can be determined.

[0076] If there are multiple objects in the priority area, the priority determination unit 15 can determine the priority of the objects based on their distance from the center point of the priority area AA. That is, the object closest to the center point of the priority area AA will be the object of attention with the highest priority, and the priority can be lower as the distance from the center point increases. The order of steps S106 and S107 is not particularly limited.

[0077] Returning to FIG. 3, parameter setting unit 16 determines image quality adjustment parameters for the input image to be subjected to recognition processing so as to improve the recognition rate, which indicates the likelihood of recognition of high-priority objects in image recognition device 30.

[0078] Specifically, the parameter setting unit 16 first calculates a control recognition rate that serves as the basis for feedback control of the image quality adjustment parameters, taking into account the priority order. That is, the image quality adjustment parameters function as a control recognition rate calculation unit that calculates a control recognition rate using the recognition rate of each object, taking into account the priority order. The control recognition rate calculation method can be optimized depending on the system, purpose of use, detection environment, etc. For example, the following calculation methods (1) to (3) can be considered as methods for calculating the control recognition rate. (1) Use the recognition rate of the object with the highest priority. (2) Among the multiple detected objects, the average recognition rate of the objects in the top 20% of priority is used. (3) The recognition rate is calculated by taking into account the weighting factor for each priority.

[0079] In the method (3), for example, weighting coefficients whose sum total is 1 are assigned so that the coefficients with larger values ​​are assigned to objects with higher priorities. The control recognition rate can be the sum of the products of the recognition rates of each object and the assigned weighting coefficients.

[0080] The parameter setting unit 16 determines the image quality adjustment parameters of the recognition signal processing unit 24 so that the control recognition rate is a high value. FIG. 12 shows an example of how the recognition rate changes with changes in the image quality adjustment parameters. In the example shown in FIG. 12, the image quality adjustment parameter is the luminance gain (digital gain) in the recognition signal processing unit 24. As shown in FIG. 12, the recognition rate increases up to a value Y1 where the luminance gain is small and the image data of the recognition target is dark for the recognition process. Thereafter, the recognition rate remains approximately constant between values ​​Y1 and Y2, which do not significantly affect the recognition process, and decreases when the image data is too bright for the recognition process and is greater than Y2. In other words, when the luminance gain is changed, the recognition rate graph exhibits a flat-topped mountain-like characteristic.

[0081] As described above, image recognition support device 10 stores and manages the recognition rate and image quality adjustment history for each object in storage unit 2. In other words, storage unit 2 can also be considered a management unit that manages changes in the recognition rate of high-priority objects when image quality adjustment parameters are changed in recognition signal processing unit 24. Parameter setting unit 16 determines changes in the control recognition rate following the previous change in image quality adjustment parameters, and, by referring to the changes in the control recognition rate, can determine new image quality adjustment parameters that will improve the control recognition rate.

[0082] Specifically, when the control recognition rate has improved, the parameter setting unit 16 changes the new image quality adjustment parameter in an increasing direction if the image quality adjustment parameter was increased during the previous adjustment of the image quality adjustment parameter, or in a decreasing direction if the image quality adjustment parameter was decreased during the previous adjustment of the image quality adjustment parameter. In other words, when the control recognition rate has improved, the parameter setting unit 16 changes the image quality adjustment parameter in the same direction as during the previous adjustment.

[0083] Furthermore, if the control recognition rate is decreasing, the parameter setting unit 16 changes the new image quality adjustment parameter in a decreasing direction if the image quality adjustment parameter was increased during the previous adjustment of the image quality adjustment parameter, and in an increasing direction if the image quality adjustment parameter was decreased. In other words, if the control recognition rate is decreasing, the parameter setting unit 16 changes the image quality adjustment parameter in the opposite direction to that during the previous adjustment.

[0084] For example, in the region below the value Y1 shown in FIG. 12, increasing the brightness gain increases the recognition rate. In this case, the recognition rate can be further improved by further increasing the brightness gain. Also, in the region above the value Y2 shown in FIG. 12, decreasing the brightness gain increases the recognition rate. In this case, the recognition rate can be further improved by further decreasing the brightness gain.

[0085] In this way, by repeatedly readjusting the image quality adjustment parameters, parameter setting unit 16 can determine the image quality adjustment parameters so that the recognition rate approaches the flat peak shown in Fig. 12. Parameter setting unit 16 can set the determined image quality adjustment parameters in recognition signal processing unit 24 via IF unit 3. This enables recognition signal processing unit 24 to generate recognition image data that will result in a high recognition rate in the recognition process by image recognition device 30.

[0086] Note that, since image recognition support device 10 performs feedback control of image quality adjustment parameters based on the control recognition rate, the image quality adjustment parameters will always fluctuate near the peak of the recognition rate. To avoid such fluctuations in image quality adjustment parameters, parameter setting unit 16 may perform control such as not changing the image quality adjustment parameters once the control recognition rate reaches a value near the peak in Fig. 12 unless the control recognition rate falls below a predetermined threshold (hereinafter referred to as the readjustment threshold).

[0087] When determining the priority of an object using an extension line of the object's velocity vector, the image recognition support device 10 may also take into account the line of sight of the object to comprehensively determine the priority. Fig. 13 is a diagram illustrating the process of determining the priority by taking into account the line of sight of the object.

[0088] 13, in the same manner as described above, the recognition signal processing unit 24 generates image data for recognition using the initial setting image quality adjustment parameters and outputs the data to the image recognition device 30. The image recognition device 30 performs recognition processing on the input image data for recognition and generates a recognition result.

[0089] First, the image recognition support device 10 acquires the recognition result obtained by performing image recognition processing of a plurality of objects on the image for recognition by the image recognition device 30, and acquires the position information of the plurality of objects on the image for recognition based on the acquired recognition result. Then, the image recognition support device 10 calculates the moving direction and speed (velocity vector) of the objects from the change in the position information (step S201).

[0090] Then, a process for determining the speed of each object is performed (step S202). A first reference value and a second reference value are used in this process. The first reference value is a threshold value for distinguishing between a person moving "normally" and a person moving "at high speed" as described above. The first reference value can be set to, for example, 6 km / h.

[0091] Normally, a moving person is likely to look in the direction of travel, but a person who is nearly stationary is likely to direct their gaze toward an object of interest. In this example, the priority of the object is determined taking into account the gaze of a nearly stationary person. The second reference value is a threshold value for distinguishing between a moving person and a person who is barely moving (hereinafter referred to as "almost stationary"). The second reference value can be, for example, 1 km / h.

[0092] If the speed of the object is less than the second reference value, the object is "substantially stationary." In this case, the process proceeds to step S203. In step S203, the gaze direction is calculated from the face orientation information. If the speed of the object is equal to or greater than the second reference value and less than the first reference value, the object is "moving normally." In this case, the process proceeds to step S205.

[0093] If the speed of the object is equal to or greater than the first reference value, the object is moving at a high speed. In this case, the process proceeds to step S204. In step S204, an extension line of the velocity vector of the object is set.

[0094] In step S205, it is determined whether or not processing has been completed for all objects included in the recognition image. If processing has not been completed for all objects (step S205, NO), the process returns to step S201, and the same processing is repeated. If processing has been completed for all objects (step S205, YES), a process of identifying priorities is executed using the intersections of the extension lines of the velocity vectors (step S206). This process is the same as steps S104 to S109 described above, and a description thereof will be omitted.

[0095] Then, it is determined whether or not there is an object with the highest priority (object of interest) (step S207). If there is no object of interest (step S207, NO), the objects are not assigned a priority and the process ends (step S208). If there is an object of interest (step S207, YES), a gaze concentration position is calculated from the gaze direction calculated in S203 (step S209). The gaze concentration position refers to an attention area that is being watched by objects that are substantially stationary.

[0096] Then, it is determined whether or not there is an object on which lines are concentrated within the priority area (step S210). If there is no object on which lines are concentrated within the priority area (step S210, NO), the priority of each object is not assigned and the process ends (step S208). If there is an object on which lines are concentrated within the priority area (step S210, YES), a final priority is determined (step S211). For example, the final priority is determined so that, of the multiple objects present within the priority area, the object on which lines are concentrated is given the highest priority.

[0097] For example, in the case of an event that does not attract much attention from people around, such as a minor collision between a bicycle and a person, people far from the location where the event occurred may not take any action but may simply turn their faces or bodies in that direction.By estimating the area of ​​interest based on the gazes of multiple people and giving higher priority to people whose gazes are concentrated within the area of ​​interest, it becomes possible to monitor even more detailed and special situations without missing them.

[0098] Here, a method for setting image quality adjustment parameters that is executed after the above-mentioned priority determination process will be described with reference to Fig. 14. Fig. 14 is a flow diagram showing the flow of processing for setting image quality adjustment parameters in image recognition assistance processing according to an embodiment. Fig. 14 also shows the flow of feedback control of image quality adjustment parameters that is repeatedly executed until the recognition rate of the target object in the recognition image stabilizes at a high level.

[0099] As described above, the recognition rate and image quality adjustment history for each object are managed in the storage unit 2. The image recognition assistance device 10 can refer to the previous recognition rate for the target object and the previous image quality adjustment parameters. First, a control recognition rate is calculated taking into account the priority order identified in the process of either FIG. 11 or FIG. 13 (step S10). The control recognition rate can be calculated, for example, taking into account a weighting coefficient for each priority order. Then, the image recognition assistance device 10 compares the current control recognition rate with the previous control recognition rate (S11).

[0100] If the current control recognition rate is lower than the previous control recognition rate, the process proceeds to step S12. In step S12, the image recognition assistance device 10 determines whether the current control recognition rate is near the peak. If the control recognition rate is near the peak (YES in S12), it is determined whether the control recognition rate is equal to or lower than the readjustment threshold (S13). If the control recognition rate is not near the peak (NO in S12), or if the control recognition rate is near the peak but is equal to or lower than the readjustment threshold (YES in S13), the image quality adjustment parameter is changed in the opposite direction to the direction of change of the previous image quality adjustment parameter (S14).

[0101] Then, image recognition support device 10 saves the current object position, recognition rate, and changed image quality adjustment parameters (S15). On the other hand, if the control recognition rate is greater than the readjustment threshold (S13, NO), image recognition support device 10 saves the current object position, recognition rate, and image quality adjustment parameters before change without changing the image quality adjustment parameters (S15).

[0102] If the current control recognition rate is higher than the previous control recognition rate, the process proceeds to step S20. In step S20, the image recognition assistance device 10 determines whether the current control recognition rate is near the peak. If the control recognition rate is near the peak (YES in S20), it is determined whether the control recognition rate is equal to or lower than the readjustment threshold (S21). If the control recognition rate is not near the peak (NO in S20), or if the control recognition rate is near the peak but is equal to or lower than the readjustment threshold (YES in S21), the image quality adjustment parameter is changed in the same direction as the direction of change in the previous image quality adjustment parameter (S22).

[0103] Then, image recognition support device 10 saves the current object position, recognition rate, and changed image quality adjustment parameters (S15). On the other hand, if the control recognition rate is greater than the readjustment threshold (S21, NO), image recognition support device 10 saves the current object position, recognition rate, and image quality adjustment parameters before change without changing the image quality adjustment parameters (S15).

[0104] If it is determined in S11 that the current control recognition rate has not changed from the previous control recognition rate, the process proceeds to step S15, where the current object position, recognition rate, and image quality adjustment parameters before the change are saved.

[0105] As described above, according to the embodiment, the recognition results from the image recognition device 30 are used to determine the priority of objects based on the behavior patterns of multiple objects in the recognition image. Then, by adjusting the image quality of the recognition image input to the image recognition device 30 so that the recognition rate of high-priority objects is high, it becomes possible to support improvement of recognition accuracy in line with the detection purpose. For example, when multiple objects are included in the recognition image, it becomes possible to improve the recognition accuracy of high-priority objects, such as objects of interest such as dangerous individuals, among these objects.

[0106] Each functional block shown in the drawings that performs various processes can be configured in hardware with a processor, memory, and other circuits. The processes described above can also be implemented by having a processor execute a program. Therefore, these functional blocks can be implemented in various forms, such as hardware only, software only, or a combination of both, and are not limited to any one of these.

[0107] The above-described program can be stored and supplied to a computer using various types of non-transitory computer-readable media. Non-transitory computer-readable media include various types of tangible storage media. Examples of non-transitory computer-readable media include semiconductor memory (e.g., mask ROM, PROM (Programmable ROM), EPROM (Erasable PROM), flash ROM, and RAM). The program may also be supplied to a computer by various types of transitory computer-readable media. Examples of transitory computer-readable media include electrical signals, optical signals, and electromagnetic waves. The transitory computer-readable medium can supply the program to a computer via a wired communication path such as an electric wire or optical fiber, or via a wireless communication path.

[0108] In the above embodiment, the recognition image with a high recognition rate is generated by adjusting the image quality of the image data captured by the camera unit 21 using software calculations in the recognition signal processing unit 24, but this is not limiting. For example, it is also possible to adjust the brightness of the recognition image by adjusting the amount of light incident on the image sensor using a mechanical shutter or iris diaphragm of the image capture device 20. Specifically, the image capture device 20 may control the mechanical shutter to adjust the charge accumulation time of the image sensor in accordance with the exposure time per frame specified by the image quality adjustment parameters input from the image recognition support device 10.

[0109] The contents of the present disclosure can be used in various fields that utilize image recognition. Some or all of the above embodiments can be described as, but are not limited to, the following supplementary notes.

[0110] (Appendix A1) a recognition result acquisition unit that acquires a recognition result obtained by performing a recognition process on an input image of a plurality of objects using an image recognition device on an input image input from an imaging device that includes an image quality adjustment unit that adjusts image quality using image quality adjustment parameters; a position information acquisition unit that acquires position information of the plurality of objects on the input image based on the recognition result; an extension line setting unit that calculates velocity vectors of the plurality of objects and sets extension lines of the respective velocity vectors; a priority area specifying unit that specifies a priority area based on intersections of the set extension lines; a priority order determination unit that determines a priority order of the object based on the identified priority area; a parameter setting unit that determines image quality adjustment parameters for an input image to be subjected to recognition processing so as to improve a recognition rate indicating the likelihood of recognition results for objects with high priority in the image recognition device, and sets the determined image quality adjustment parameters in the image quality adjustment unit; an attention area estimation unit that estimates an attention area that the object is paying attention to based on information about the object's line of sight; Including, the priority order determination unit determines the priority order in consideration of the region of interest. Image recognition support device. (Appendix A2) The attention area estimation unit estimates the attention area based on information about a line of sight of an object that is substantially stationary. 10. The image recognition assistance device according to claim 1. (Appendix B1) The computer a process of acquiring a recognition result in which an input image input from an imaging device having an image quality adjustment unit that adjusts image quality using image quality adjustment parameters is subjected to a recognition process for a plurality of objects by an image recognition device; A process of acquiring position information of the plurality of objects on the input image based on the recognition result; calculating a velocity vector of each of the plurality of objects and setting an extension line of each of the velocity vectors; A process of identifying a priority area based on intersections of the set extension lines; determining a priority of the object based on the identified priority area; determining image quality adjustment parameters for an input image to be subjected to recognition processing so as to improve a recognition rate indicating the likelihood of recognition results for objects with high priority in the image recognition device, and setting the determined image quality adjustment parameters in the image quality adjustment unit; A method for assisting image recognition. [Explanation of symbols]

[0111] 100 Image Recognition System 1 Processing section 2 Storage section 3 IF Section 10 Image recognition support device 11 Recognition result acquisition section 12 Location information acquisition unit 13 Extension line setting section 14 Priority area identification part 15 Priority Determination Unit 16 Parameter setting section 17. Region of interest estimation unit 20 Imaging device 21 Camera unit 22 Signal processing section 23 Image output unit 24 Recognition signal processing section 30 Image recognition device 40 Display device 50 Input Device P0 Recognition image P1~P8 People AA priority area

Claims

1. a recognition result acquisition unit that acquires a recognition result obtained by performing a recognition process on an input image of a plurality of objects using an image recognition device on an input image input from an imaging device that includes an image quality adjustment unit that adjusts image quality using image quality adjustment parameters; a position information acquisition unit that acquires position information of the plurality of objects on the input image based on the recognition result; an extension line setting unit that calculates velocity vectors of the plurality of objects, and sets extension lines of the respective velocity vectors based on the magnitudes of the velocity vectors; a priority area specifying unit that specifies a priority area based on intersections of the set extension lines; a priority order determination unit that determines a priority order of the object based on the identified priority area; a parameter setting unit that determines image quality adjustment parameters for an input image to be subjected to recognition processing so as to improve a recognition rate indicating the likelihood of recognition results for objects with high priority in the image recognition device, and sets the determined image quality adjustment parameters in the image quality adjustment unit; Including, Image recognition support device.

2. the priority area specification unit extracts intersections that include at least one intersection other than the central intersection within a predetermined range centered on one of the plurality of intersections as a central intersection, and specifies a predetermined area that includes all of the extracted intersections as a priority area; the priority order determination unit assigns a higher priority to objects within the priority area than to objects outside the priority area; The image recognition support device according to claim 1 .

3. the priority order determination unit determines the priority order of the object based on the distance from the center point of the priority area. The image recognition support device according to claim 2 .

4. the priority area identification unit identifies the predetermined area as a priority area when all of the extracted intersections are intersections toward which objects approach or intersections from which objects move away. The image recognition support device according to claim 2 .

5. a process of acquiring a recognition result in which an input image input from an imaging device having an image quality adjustment unit that adjusts image quality using image quality adjustment parameters is subjected to a recognition process for a plurality of objects by an image recognition device; A process of acquiring position information of the plurality of objects on the input image based on the recognition result; calculating velocity vectors of the plurality of objects, and setting extension lines of the velocity vectors based on the magnitudes of the velocity vectors; A process of identifying a priority area based on intersections of the set extension lines; determining a priority of the object based on the identified priority area; determining image quality adjustment parameters for an input image to be subjected to recognition processing so as to improve a recognition rate indicating the likelihood of recognition results for objects with high priority in the image recognition device, and setting the determined image quality adjustment parameters in the image quality adjustment unit; A program that causes a computer to execute the following.

Images