IMAGE RECOGNITION DEVICE, IMAGE RECOGNITION DEVICE METHOD, AND IMAGE RECOGNITION DEVICE PROGRAM

Abstract

To solve a problem in which it is necessary to register the size of an object by inputting information in advance for each installation environment, which takes time and effort.SOLUTION: An image recognition device comprises: an image acquisition unit (11) that acquires an image; a detection unit (12) that detects a detection object from the image; an index creation unit (14) that creates an index for determining whether the detection object is a target object from the history of the location and size of the detection object; and a determination unit (15) that uses the index to determine whether the detection object is the target object. The index is information representing a distribution trend of the location and size of the detection object in the history.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to an image recognition device. [Background technology]

[0002] With the spread of network cameras, technologies to detect people and objects from captured images are widely used. When detecting people from captured images, objects with shapes similar to those of people or combinations of shapes in the captured images may be erroneously detected as people. If such erroneous detections occur frequently, appropriate image recognition cannot be performed.

[0003] Patent Document 1 discloses a technique for improving the detection accuracy of a target object by detecting an object of a specific size as the target object according to its distance from a reference position in a fisheye image. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2019-159739 A Summary of the Invention [Problem to be solved by the invention]

[0005] However, the above-mentioned conventional technology requires that the relationship between the size of a detected object to be determined as a person and the distance of the detected object from a reference position be registered in advance. Therefore, the user must input information about the camera's imaging environment, such as the installation height of the camera, the height of the detection object, and the relationship between the distance from the center (reference position) in the image and the distance in real space, in advance. Such input work places a heavy burden on the user.

[0006] Therefore, an object of one aspect of the present invention is to provide an image recognition device that detects a desired object with higher accuracy without requiring a user to perform complicated input operations in advance. [Means for solving the problem]

[0007] In order to solve the above problems, an image recognition device according to one embodiment of the present invention includes an image acquisition unit that acquires an captured image, a detection unit that detects a predetermined type of object from the captured image through image recognition as a detected object, a recording control unit that controls recording of a history of the position and size of the detected object in the captured image detected by the detection unit, an index creation unit that creates an index for determining whether or not the detected object is a target object to be detected based on the history, and a determination unit that determines whether or not the detected object is the target object by comparing the index with the position and size of the detected object in the captured image detected by the detection unit.

[0008] According to the above configuration, an index is created based on the history of the position and size of the detected object, and it is possible to determine whether the detected object is a target object or not based on the position and size of the detected object relative to the index. This makes it possible to improve the detection accuracy of the target object without requiring the user to take the trouble of inputting information in advance depending on the environment.

[0009] The index creating section may create, as the index, a function indicating a relationship between a position of the detected object and a size of the detected object.

[0010] According to the above configuration, a function can be derived as an index. By using the function, the size of a detected object at any position of the detected object can be uniquely derived.

[0011] The determination section may determine that the detected object is the target object when a distance determined from the function and the position and size of the detected object in the captured image is less than a predetermined threshold.

[0012] According to the above configuration, the determination unit calculates the distance between a point determined by the position and size of the detected object and a function, and can determine whether the detected object is a target object or not when the distance is less than a predetermined threshold.

[0013] The recording control unit performs control to record the correspondence between the one-dimensional position of the detected object in a predetermined axial direction in the captured image and the size of the detected object, the function being a two-dimensional approximation curve of the one-dimensional position of the detected object and the size of the detected object, and the distance may be the distance between the approximation curve and a point representing the one-dimensional position of the detected object and the size of the detected object.

[0014] According to the above configuration, it is possible to determine the size of a detected object that is a target object and to determine whether the detected object is being detected erroneously, taking into consideration only the vertical direction in the image.

[0015] The recording control unit controls recording of the correspondence between the two-dimensional position of the detected object in two predetermined axial directions in the captured image and the size of the detected object, the function being an approximation surface in three dimensions of the two-dimensional position of the detected object and the size of the detected object, and the distance may be the distance between the approximation surface and a point representing the two-dimensional position of the detected object and the size of the detected object.

[0016] According to the above configuration, it is possible to judge whether a detected object has been erroneously detected by taking into consideration the up, down, left, and right directions in the image. By using the left and right directions in the judgment, it is possible to detect the object with high accuracy even if there is a difference in height, such as a staircase, in the space being imaged.

[0017] A method of an image recognition device according to another aspect includes an image acquisition step of acquiring an image, a detection step of detecting a predetermined type of object from the imaged image by image recognition as a detected object, a recording control step of controlling recording of a history of the position and size of the detected object in the imaged image detected in the detection step, an index creation step of creating an index for determining whether or not the detected object is a target object to be detected based on the history, and a determination step of determining whether or not the detected object is the target object by comparing the position and size of the detected object in the imaged image detected in the detection step with the index.

[0018] The image recognition device according to each aspect of the present invention may be realized by a computer. In this case, the image recognition program of the image recognition device that realizes the image recognition device by a computer by causing the computer to operate as each part (software element) of the image recognition device, and the computer-readable recording medium on which it is recorded, also fall within the scope of the present invention. Effect of the Invention

[0019] According to one aspect of the present invention, it is possible to realize an image recognition device that detects a desired object more accurately without requiring a user to perform cumbersome advance input operations. [Brief description of the drawings]

[0020] [Figure 1] 1 is a block diagram showing a configuration of a main part of an image recognition system according to a first embodiment. [Diagram 2] FIG. 13 is a model diagram showing a case where a camera is attached upward and captures images in a downward direction. [Diagram 3] 11A and 11B are diagrams showing changes in captured images over time when a camera is attached above and captures images in a downward direction. [Figure 4] FIG. 13 is a model diagram showing a case where a camera is attached downward and captures images in an upward direction. [Diagram 5]11A and 11B are diagrams showing changes over time in a captured image captured with a camera attached downward and looking up. [Figure 6] This is an example of an incorrect object being detected when the camera was installed upwards. [Figure 7] This is an example of an incorrect object being detected when the camera was installed downward. [Figure 8] FIG. 13 is a diagram showing the correlation between the position (Y coordinate) and size of a detected object. [Figure 9] 4 is a flowchart showing an operation example of the image recognition system according to the first embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0021] [Embodiment 1] Hereinafter, an embodiment according to one aspect of the present invention (hereinafter, also referred to as "the present embodiment") will be described with reference to the drawings. In the drawings, the same or corresponding parts are designated by the same reference characters and their description will not be repeated.

[0022] § 1. Examples of Application 1 is a block diagram showing the configuration of the main parts of an image recognition system 100 according to embodiment 1. The image recognition system 100 includes a camera 20 and an image recognition device 1 that acquires an image captured by the camera 20 and performs image recognition. Once the camera 20 is attached, the angle of view and the imaging direction do not change and are constant.

[0023] In the image recognition device 1, a detection unit 12 detects a predetermined type of object from a captured image acquired by an image acquisition unit 11, and a determination unit 15 determines whether or not the detected object is a target object to be detected. In making the determination in the determination unit 15, it is determined whether or not the detected object is a target object based on an index acquired by a recording control unit 13.

[0024] The corresponding relationship between the position of the detected object in the captured image and the size of the target object at that position is predefined for the index, so that when the size of the detected object is close to the size of the index corresponding to the position of the detected object in the captured image, the detected object can be determined to be the target object.

[0025] §2. Example configuration (Configuration of image recognition device 1) The image recognition device 1 includes an image acquisition unit 11, a detection unit 12, a record control unit 13, a storage unit 131, an index creation unit 14, and a determination unit 15.

[0026] The image acquisition unit 11 acquires a captured image captured by the camera 20. The captured image is output to the detection unit 12.

[0027] The detection unit 12 detects a predetermined type of object in the input captured image by image recognition. A person is a typical example of the object, but the object is not limited to a person and may be any object. The position and size of the detected object, which are the detection results by the detection unit 12, are output to the recording control unit 13 and the determination unit 15.

[0028] Here, the position of the detected object is the center of a frame (detection frame) that surrounds the detected object. The method of deriving the position of the detected object is not limited to this method, and other methods such as any point on the detection frame or the center of gravity of the detected detected object may be used.

[0029] The size of the detection object may be any of the lengths of the detection frame. Examples of the length of the detection frame include the length of the diagonal, the length in the vertical direction, the length in the horizontal direction, the longer length of either the vertical direction or the horizontal direction, or the shorter length of either the vertical direction or the horizontal direction. The size of the detection object may be the area of ​​the detection frame. Furthermore, when the area occupied by the detection object itself is specified, the size of the detection object may be any of the lengths or the area of ​​that area.

[0030] The recording control unit 13 controls the data to be recorded in the memory unit 131 that records the programs and various parameters of the image recognition device 1. Specifically, the recording control unit 13 causes the memory unit 131 to record the correspondence between the position and size of the detected object as a detection history (history). The recording control unit 13 outputs the detection history recorded in the memory unit 131 to the index creation unit 14.

[0031] The storage unit 131 is not limited to a functional block of the image recognition device 1, and may be provided in an external device. That is, the recording control unit 13 may access the storage unit 131 provided in an external device via various communication means to record and acquire the detection history.

[0032] The index creation unit 14 creates an index that indicates the distribution tendency of the position and size of the detected object based on the detection history. The index is information that can be expressed in the form of a function, for example, and is used to determine whether the detected object is a target object. The index creation unit 14 outputs the created index to the determination unit 15.

[0033] The determination unit 15 determines whether or not the detected object detected by the detection unit 12 is a target object based on the index created by the index creation unit 14. The determination unit 15 outputs the determination result. The determination result by the determination unit 15 can be output to a display control unit of a display device (not shown), a communication control unit for transmitting the determination result to an arbitrary server or terminal device via a communication network, or the like.

[0034] (Image captured by installing camera 20) Fig. 2 is a model diagram showing a case where the camera 20 is attached above and captures images in a downward direction. Fig. 3 is a diagram showing changes in a captured image over time when the camera 20 is attached above and captures images in a downward direction. Fig. 4 is a model diagram showing a case where the camera 20 is attached below and captures images in an upward direction. Fig. 5 is a diagram showing changes in a captured image over time when the camera 20 is attached below and captures images in an upward direction.

[0035] The image recognition device 1 detects an object assumed to be a person from within a captured image as a detected object 32. In the following description, the origin is assumed to be at the bottom left of the captured image, the X axis is in the left-right direction, and the Y axis is in the up-down direction.

[0036] First, consider the case where camera 20 is attached upward and captures an image of object 31 looking down, as shown in Fig. 2. As shown in image 41 in Fig. 3, an object far from camera 20 appears small at the top of the captured image, and as shown in image 42 in Fig. 3, an object near camera 20 appears large at the bottom of the captured image.

[0037] As shown in FIG. 3, a detected object 32 walking from far away from camera 20 toward the vicinity of camera 20 gradually increases in size as the distance from camera 20 decreases, and its position in the captured image moves from the top (large Y coordinate) to the bottom (small Y coordinate).

[0038] Consider the case where camera 20 is attached downward and captures an image of object 31 looking up, as shown in Fig. 4. As shown in image 43 in Fig. 5, an object far from camera 20 appears small at the bottom of the captured image, and as shown in image 44 in Fig. 5, an object near camera 20 appears large at the top of the captured image.

[0039] As shown in FIG. 4, a detected object 32 walking from far away from camera 20 to near camera 20 gradually increases in size as the distance from camera 20 decreases, and its position in the captured image moves from downward (small Y coordinate) to upward (large Y coordinate).

[0040] That is, regardless of how camera 20 is attached, the position and size of the detected object changes in accordance with the movement of the detected object.

[0041] (False positive by camera 20) Fig. 6 shows an example in which the wrong object is detected when the camera 20 is attached upward. Fig. 7 shows an example in which the wrong object is detected when the camera 20 is attached downward.

[0042] 6, when the camera 20 is attached upward, the image recognition device 1 may erroneously detect an object other than a person at the position shown in the detection frame 33. This occurs, for example, when the background image resembles the shape of a human body. When an object other than a person is erroneously detected, there is a high possibility that the size at the position where the detection frame 33 is displayed (Y coordinate) will be different from the size in the situation where the person described above was detected.

[0043] Similarly, when the camera 20 is attached downward as shown in FIG. 7, the image recognition device 1 may erroneously detect an object that is not a person at the position shown in the detection frame 33.

[0044] That is, in FIG. 6, when camera 20 is attached upward, detection frame 33 is displayed below the image. Since detection frame 33 is actually located at the bottom of the image, the detection frame should be relatively large, but the detected detection frame 33 is relatively small. Similarly, in FIG. 7, when camera 20 is attached downward, detection frame 33 is displayed above the image. Since detection frame 33 is actually located at the top of the image, the detection frame should be relatively large, but the detected detection frame 33 is relatively small. In this way, the position and size of the detection frame when a false detection occurs is likely to be different from when a person is detected. Therefore, it is possible to determine a false detection based on an index indicating the position and size of the detection frame.

[0045] (Correlation between the position and size of the detected object 32) FIG. 8 is a diagram showing the correlation between the position (Y coordinate) and size of detected object 32. A detection history that associates the position and size of previously detected objects is recorded in memory unit 131. FIG. 8 shows this detection history plotted on a two-axis scatter diagram as a detection history distribution. In this detection history distribution, the horizontal axis represents the size of the detected object, and the vertical axis represents the position of the detected object (position in the vertical direction of the captured image (Y coordinate)). The numerous points 51 show a certain degree of correlation, forming a discrete distribution.

[0046] The index creation unit 14 acquires the detection history from the storage unit 131, and derives an approximation curve 52 based on a large number of points 51. The approximation curve 52 can be derived, for example, by the least squares method, but is not limited thereto, and any approximation curve derivation method may be used.

[0047] Here, when the detection history includes erroneous detection data at the stage of creating the detection history, it is considered that there will be points that deviate significantly from the approximation curve. However, when the number of such deviating points is small compared to the others, the impact on the approximation curve 52 is small and can be ignored. In other words, when the probability of erroneous detection is low, it is possible to create a sufficiently appropriate index even if erroneous detection data is included at the stage of creating the detection history. After the number of detection histories reaches a certain number or more, only data in which the determination unit 15 determines that the detected object is the target object is added to the detection history, thereby eliminating the adverse effect of erroneous detection data on index creation.

[0048] The index creating unit 14 creates an approximation curve 52 that indicates the distribution tendency of the multiple points 51, but is not limited to this and may be any index. For example, a function may be used. The approximation curve 52 that represents the function is not limited to a continuous curve, but may be a discontinuous curve or straight line for each predetermined range of position and size. The function may also be expressed in a table format in which a value is determined for each predetermined range of position and size.

[0049] Here, the determination process in the determination unit 15 in a state where the above-mentioned index is obtained will be described. When the detected object 32 is detected, the determination unit 15 compares the position and size of the detected object 32 with the above-mentioned function as the index. Conceptually, this comparison process involves plotting the position and size of the detected object 32 at point 53 shown in FIG. 8, calculating distance 54 from approximation curve 52, and comparing it with a predetermined value. Here, distance 54 is the shortest distance between point 53 and approximation curve 52. Note that distance 54 may be the difference in Y coordinate between point 53 and approximation curve 52. The calculation method of distance 54 is not limited to these, and any calculation method may be used.

[0050] §3. Example of operation FIG. 9 is a flowchart showing an operation example of the image recognition system 100 according to the first embodiment.

[0051] In S11, the image acquisition unit 11 acquires a captured image from the camera 20. The acquired captured image is output to the detection unit 12. The detection unit 12 detects an object 31 from the input captured image, and sets it as a detected object 32.

[0052] In S12, the detection unit 12 determines whether or not it has detected at least one detection object 32. If it has not detected at least one detection object 32 (No in S12), it proceeds to S13. On the other hand, if it has detected at least one detection object 32 (Yes in S12), it proceeds to S14.

[0053] In S13, the determination unit 15 indicates that the detected object 32 is not present.

[0054] In S14, the recording control unit 13 determines whether the number of detection histories is equal to or greater than a predetermined number. If the number is less than the predetermined number (No in S14), the process proceeds to S19. On the other hand, if the number is equal to or greater than the predetermined number (Yes in S14), the process proceeds to S15.

[0055] In S15, the index creating unit 14 reads out the detection history from the storage unit 131, and derives, as an index, an approximation curve 52 that reflects the tendency of the detection history distribution.

[0056] In S16, a distance 54 between the approximation curve 52 and a point 53 corresponding to the position and size of the currently detected object 32 is calculated. If the calculated distance 54 is equal to or greater than a predetermined threshold (No in S16), the process proceeds to S17. On the other hand, if the calculated distance 54 is less than the predetermined threshold (Yes in S16), the process proceeds to S18.

[0057] In S17, the determination unit 15 determines that the detected object 32 is not a person (target object) and is an erroneous detection.

[0058] In S18, the determination unit 15 determines that the detected object 32 is a person.

[0059] In S19, the record control unit 13 stores the position and size of the detected object in the detection history.

[0060] In S20, the determination unit 15 outputs the results of the detection unit 12 and the determination unit 15, and displays them on the display device.

[0061] Here, the user can visually check the results of the detection unit 12 and the determination unit 15, and for detected objects that are clearly considered to be erroneous detections or erroneous determinations, the recording control unit 13 can set a flag indicating erroneous detection so that the detected objects are not used in generating the approximation curve for the detection history distribution. This operation improves the accuracy of determining the target object among the detected objects.

[0062] § 4. Actions and Effects Whether the camera 20 is mounted above and captures an image looking down, or whether the camera 20 is mounted below and captures an image looking up, the position and size of the detected object on the image will change as the detected object moves. Since there is a relationship between the change in position and size of the detected object, an index is derived based on the distribution trend obtained from the relationship. By comparing the position and size of the detected object with the index, it is possible to determine whether the detected object is the target object that was intended to be detected. Therefore, it is possible to determine whether an object that is not the target object has been erroneously detected, and the detection accuracy of the target object can be improved.

[0063] Therefore, the position and size of a detected object are automatically recorded, and when the number of combinations of recorded position and size exceeds a predetermined number, it becomes possible to determine whether the detected object is a target object (e.g., a person) or another object. Therefore, it is easy to use because it becomes possible to detect and determine a person without inputting information such as the size of the person in advance for each installation environment according to the mounting position of the camera 20.

[0064] By collecting a predetermined number of histories of the positions and sizes of detected objects, the distribution trend of the positions and sizes of detected objects can be understood, and a highly accurate index can be created.

[0065] In particular, the index may be a function, or even an approximation curve in a two-dimensional space, and it is possible to determine whether or not a detected object is a target object by taking into account only the up and down directions in an image.

[0066] In addition, the approximation curve is considered as the size of the detected object in the vertical direction, but is not limited to this, and may be the size of the detected object in a one-dimensional position in any one-axis direction. That is, the approximation curve may be the size of the detected object in the horizontal direction, or the size of the detected object in a line with any inclination on the captured image, and the approximation curve may be a function obtained from a point group representing the one-dimensional position and the size of the detected object. For example, when the camera 20 is installed on the side wall of the corridor at approximately the same height as the human body, and the imaging direction is determined so that the far side of the corridor is on the left side of the captured image and the near side of the corridor is on the right side of the captured image, the size of the target object increases from the left side to the right side of the captured image. In this case, the approximation curve may be set according to the size of the detected object in the horizontal direction of the captured image.

[0067] [Variation 1] In the first embodiment, only the vertical direction of the image is considered for the position of the detected object, but in the first modification, an example will be described in which the vertical, horizontal and vertical directions of the image are also considered. That is, the size of the detected object is associated with the X and Y coordinates on the image.

[0068] Therefore, the detection history distribution is obtained as a scatter diagram on three axes: the X coordinate of the detected object, the Y coordinate of the detected object, and the size of the detected object. In addition, the index derived from the detection history distribution is an approximated surface. The approximated surface may be expressed by a function. In addition, the approximated surface representing the function is not limited to a continuous surface, but may be a discontinuous surface or a plane for each predetermined range of position and size. In addition, the function may be expressed in a table format in which a value is determined for each predetermined range of position and size.

[0069] Therefore, in the first modification, the distance of a new point that represents the X-coordinate, Y-coordinate, and size of a detected object in three dimensions to the approximation surface can be calculated. The distance may be calculated by calculating the distance to a point on the approximation surface that is closest to the new point. Alternatively, the distance may be calculated by calculating the distance to a point on the approximation surface that corresponds to the size of the new point in terms of the X-coordinate and Y-coordinate. If these distances are less than a predetermined value, the detected object is determined to be a target object (person).

[0070] In the first modification of the first embodiment, the detection history distribution can be expressed as a three-axis scatter diagram, so that an approximated surface can be created that takes into account the relationship of three-dimensional depth. Therefore, even if the Y coordinate is the same, if the X coordinate is different, it can handle cases where the size is different.

[0071] For example, if the space being imaged has a difference in height, such as a staircase, the size of the target object may change differently depending on the position in the image. In such a case, by setting an approximate curved surface according to the difference in height, it becomes possible to accurately determine whether or not the target object is present.

[0072] Furthermore, when an area within the captured image where an object such as a wall cannot enter is included, the area where the object cannot exist can be clearly defined by setting the size of the detected object to 0 or a maximum value in advance for the wall area, etc. In this case, however, the user must input information in advance that indicates the relationship between the position and size of the detected object.

[0073] In addition, the approximated surface is considered in terms of the size of the detected object relative to the two-dimensional space of the X and Y coordinates, but is not limited to this and may be considered in terms of the size of the detected object relative to the two-dimensional position in two predetermined axial directions. In this case, the function may be a three-dimensional approximated surface of the two-dimensional position of the detected object and the size of the detected object.

[0074] [Modification 2] In the first embodiment, the camera 20 is angled with respect to the vertical direction, but in the second modification, a case where the optical axis is oriented vertically will be described. For example, the camera 20 is attached to the ceiling facing straight down.

[0075] In such cases, it is common to use a fisheye lens or a wide-angle lens, etc. With these lenses, even if they are equidistant from the camera 20, an object that is angled with respect to the optical axis of the camera 20 tends to appear smaller in the captured image than an object that is on the optical axis.

[0076] In addition, the captured image has the property that the images are arranged concentrically around the optical axis. Therefore, unlike the tendency in the first embodiment, the detection history shows the position of the detected object not as a Y coordinate but as a radius from the center position.

[0077] [Modification 3] In the first embodiment, whether or not a detected object is a target object is determined based on the relationship between the approximation curve and the position and size of the detected object. In the third modification, whether or not a detected object is a target object is determined by AI (Artificial Intelligence) based on the detection history distribution and the position and size of the detected object without using the approximation curve. For example, machine learning for outlier determination may be used.

[0078] Even in this case, even if an obvious outlier corresponding to a false positive is mixed into the detection history distribution, the influence of the outlier is minimized in the learning stage and becomes negligible.

[0079] [Software implementation example] The functions of the image recognition device 1 (hereinafter referred to as the "device") can be realized by a program for causing a computer to function as the device, and by a program for causing a computer to function as each control block of the device (in particular, each part included in the image recognition device 1).

[0080] In this case, the device includes a computer having at least one control device (e.g., a processor) and at least one storage device (e.g., a memory) as hardware for executing the program. The control device and storage device execute the program to realize each function described in each of the above embodiments.

[0081] The program may be non-transitory and may be recorded in one or more computer-readable recording media. The recording media may or may not be included in the device. In the latter case, the program may be provided to the device via any wired or wireless transmission medium.

[0082] In addition, some or all of the functions of each of the control blocks can be realized by a logic circuit. For example, an integrated circuit in which a logic circuit that functions as each of the control blocks is formed is also included in the scope of the present invention. In addition, the functions of each of the control blocks can be realized by, for example, a quantum computer.

[0083] [Additional Notes] The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the present invention. [Explanation of symbols]

[0084] 1. Image recognition device 11 Image acquisition section 12 Detection section 13 Storage section 14. Index Creation Department 15 Judgment section 20 Camera 31 Object 32 Detected object 33 False positive frame 51 Many points 52 Approximate curve (indicator) 53 New points 54 distance 100 Image Recognition System

Claims

1. an image acquisition unit that acquires a captured image; a detection unit that detects a predetermined type of object as a detection object from the captured image by image recognition; a record control unit that controls recording of a history of a position and a size of the detected object in the captured image, the detected object being detected by the detection unit; an index creating unit that creates an index for determining whether the detected object is a target object to be detected based on the history; and a determination unit that determines whether the detected object is the target object by comparing the position and size of the detected object in the captured image detected by the detection unit with the index.

2. The image recognition device according to claim 1 , wherein the index generating unit generates, as the index, a function indicating a relationship between a position of the detected object and a size of the detected object.

3. The image recognition device according to claim 2 , wherein the determination unit determines that the detected object is the target object when a distance determined from the function and the position and size of the detected object in the captured image is less than a predetermined threshold value.

4. the recording control unit performs control to record a correspondence relationship between a one-dimensional position of the detected object in a predetermined axis direction in the captured image and a size of the detected object, the function is a two-dimensional approximation curve of the one-dimensional position of the detected object and the size of the detected object, The image recognition device according to claim 3 , wherein the distance is a distance between the approximation curve and a point representing the one-dimensional position and size of the detected object.

5. the recording control unit performs control to record a correspondence relationship between a two-dimensional position of the detected object in two predetermined axial directions in the captured image and a size of the detected object, the function is a three-dimensional approximation surface of the two-dimensional position of the detected object and the size of the detected object, 4. The image recognition device according to claim 3, wherein the distance is a distance between the approximate curved surface and a point representing the two-dimensional position and size of the detected object.

6. an image acquisition step of acquiring a captured image; a detection step of detecting a predetermined type of object as a detection object from the captured image by image recognition; a recording control step of controlling recording of a history of the position and size of the detected object in the captured image, the position and size being detected in the detection step; an index creating step of creating an index for determining whether the detected object is a target object to be detected based on the history; A method for an image recognition device including a determination step of determining whether the detected object is the target object by comparing the position and size of the detected object in the captured image detected in the detection step with the index.

7. An image recognition program for causing a computer to function as the image recognition device described in claim 1, comprising the image recognition program for causing a computer to function as the image acquisition unit, the detection unit, the recording control unit, the index creation unit, and the judgment unit.

Images