Recognition processing device, recognition processing method, and recognition processing program

Abstract

To provide a technique for more accurately detecting an object in image recognition processing.SOLUTION: A recognition processing device 10 includes an image acquisition unit 11 that acquires an image, a recognition unit 12 that recognizes an object from the image, and a correction unit 13 that corrects a range of the object on the basis of an attenuation mode of luminance in a vertical direction of a frame of the image in the range of the recognized object.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a recognition processing device, a recognition processing method, and a recognition processing program. [Background technology]

[0002] There is known a technology for detecting objects such as pedestrians from an image captured around a vehicle using image recognition processing such as pattern matching. For example, Patent Document 1 discloses a technology for detecting with high accuracy the range in which an object actually exists when the ground on which the object exists is wet and the reflected image of the object reflected on the ground is visible.

[0003] The technology described in Patent Document 1 estimates the distance from the imaging device to the object by detecting upper and lower symmetrical parts at the bottom of the range of the object recognized from the image generated by the imaging device, and correcting the range of the object so that the symmetrical center position of the upper and lower symmetrical parts is the lower end of the detection range of the object. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Publication No. 2022-15696 Summary of the Invention [Problem to be solved by the invention]

[0005] In reality, the image of an object reflected by the road surface may not be symmetrical from top to bottom. In this case, the technology described in Patent Document 1 may not properly detect the symmetrical parts from top to bottom, and may not properly recognize the range of the object.

[0006] The present disclosure has been made in consideration of the above circumstances, and aims to provide a technology for recognizing objects with higher accuracy in image recognition processing. [Means for solving the problem]

[0007] A recognition processing device according to one aspect of the present disclosure includes an image acquisition unit that acquires an image, a recognition unit that recognizes an object from the image, and a correction unit that corrects the range of the recognized object based on the brightness attenuation pattern in the vertical direction of the frame of the image within the range of the recognized object.

[0008] A recognition processing method according to another aspect of the present disclosure includes the steps of acquiring an image, recognizing an object from the image, and modifying the range of the recognized object based on the manner in which brightness attenuates in the vertical direction of the frame of the image within the range of the recognized object.

[0009] A recognition processing program according to yet another aspect of the present disclosure causes a computer to perform the steps of acquiring an image, recognizing an object from the image, and modifying the range of the recognized object based on the brightness attenuation pattern in the vertical direction of the frame of the image within the range of the recognized object. [Effects of the Invention]

[0010] According to an aspect of the present disclosure, a technique for recognizing objects with higher accuracy in image recognition processing can be provided. [Brief explanation of the drawings]

[0011] [Figure 1] 1 is a block diagram illustrating a configuration of a recognition device according to a first embodiment. [Figure 2] FIG. 1 is a diagram schematically illustrating an example of an image including an object. [Figure 3] FIG. 3 is a diagram illustrating an example of brightness values ​​within the recognition frame of FIG. 2. [Figure 4] 3 is a diagram illustrating an example of gradient values ​​of brightness within the recognition frame of FIG. 2. FIG. [Figure 5] FIG. 10 is a diagram illustrating a reflective boundary. [Figure 6]FIG. 10 is a diagram schematically illustrating an example of an image after the recognition frame has been corrected. [Figure 7] 4 is a flowchart showing an example of processing in a recognition processing method according to the first embodiment. [Figure 8] FIG. 10 is a block diagram illustrating the configuration of a recognition device according to a second embodiment. [Figure 9] FIG. 10 is a diagram illustrating an example of an estimation result of a skeleton of a person in an image. [Figure 10] 10 is a flowchart showing an example of processing in a recognition processing method according to a second embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0012] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. Specific numerical values ​​and the like shown in the embodiments are merely examples for facilitating understanding of the invention, and do not limit the present disclosure unless otherwise specified. Elements not directly related to the present disclosure are omitted from the drawings.

[0013] First embodiment FIG. 1 is a block diagram schematically illustrating the configuration of a recognition device 1 according to a first embodiment. The recognition device 1 according to the first embodiment includes a recognition processing device 10, a camera 21, and an output device 30. Each functional block of the recognition processing device 10 can be realized, for example, by a combination of hardware and software. The hardware of the recognition processing device 10 is realized by elements and mechanical devices, such as a processor such as a central processing unit (CPU) or a graphics processing unit (GPU) of a computer, and memories such as read-only memory (ROM) or random access memory (RAM). The software of the recognition processing device 10 is realized by a computer program or the like. The recognition processing device 10 includes, as functional blocks realized by a combination of hardware and software, an image acquisition unit 11, a recognition unit 12, a correction unit 13, and an output control unit 14.

[0014] The recognition processing device 10 acquires an image that may include objects such as pedestrians and vehicles in the vicinity, and recognizes the objects included in the image. In the first embodiment, an example is shown in which the recognition processing device 10 is installed on a multifunctional utility pole used as road infrastructure equipment, such as a Smart Pole (registered trademark). The multifunctional utility pole of the first embodiment is installed, for example, on a street, and includes an antenna and communication equipment for providing wireless communication functions, and a camera 21 for capturing images of vehicles and pedestrians passing on the road. The recognition processing device 10 of the first embodiment may be fixed to a predetermined location, like a multifunctional utility pole, or may be mounted on a moving object or flying object, such as a vehicle or drone.

[0015] The camera 21 is provided on the multifunction utility pole and generates an image capturing the surroundings of the multifunction utility pole. The camera 21 is provided, for example, above the multifunction utility pole and generates an image with an angle of view looking down on the ground on which the multifunction utility pole is installed. The camera 21 of the first embodiment is a far-infrared camera that captures infrared light to generate a thermal image. However, the camera 21 of the first embodiment is not limited to this, and may be, for example, a visible light camera that captures visible light to generate a color image or a monochrome image. The image generated by the camera 21 is, for example, a moving image at 30 frames per second or 60 frames per second, but may also be a still image.

[0016] The output device 30 is a communication device that transmits the results of object recognition by the recognition processing device 10 to another device. The output device 30 transmits the object recognition results to, for example, a server used by a business operator that manages a multifunctional utility pole. The business operator that manages the multifunctional utility pole transmits the object recognition results to a device mounted on a vehicle traveling around the multifunctional utility pole or an information terminal carried by a person near the multifunctional utility pole. The output device 30 may be a communication device that transmits the object recognition results to a device mounted on a vehicle traveling around the multifunctional utility pole or an information terminal carried by a person near the multifunctional utility pole. The device mounted on a vehicle traveling around the multifunctional utility pole or the information terminal carried by a person near the multifunctional utility pole may be, for example, a vehicle infotainment system, a car navigation device, a PDA (Personal Digital Assistant), a mobile phone, a smartphone, a tablet terminal, etc.

[0017] The image acquisition unit 11 acquires an image generated by the camera 21. The image acquisition unit 11 may be an imaging control device that controls the camera 21.

[0018] The recognition unit 12 recognizes objects in the images acquired by the image acquisition unit 11. The recognition unit 12 recognizes objects in the images using a known object recognition method, such as YOLO (YOU Only Look Once), which uses a model trained on images of various objects to be recognized. The recognition unit 12 recognizes objects such as people, automobiles, bicycles and motorcycles with people on them, for example. When an object is recognized in the image, in other words, when an object captured in the image is detected, the recognition unit 12 assigns a rectangular recognition frame surrounding the recognized object. The recognition frame in the first embodiment is an example of the range of the object.

[0019] The correction unit 13 corrects the recognition frame based on the brightness attenuation pattern in the recognition frame. The correction unit 13 includes an identification unit 13A. The correction unit 13 may also include a mask processing unit 13B in addition to the identification unit 13A. The identification unit 13A identifies a reflection boundary between a recognition target range in the recognition frame in which the object to be recognized exists and a reflection range in which a reflected image of the object to be recognized is reflected, based on the brightness attenuation pattern in the recognition frame. Specifically, the correction unit 13 identifies the reflection boundary based on a characteristic brightness attenuation characteristic that corresponds to the vertical position of the image in the recognition frame, as the brightness attenuation pattern in the recognition frame. The correction unit 13 may correct the recognition frame by changing the shape of the recognition frame so that it does not include a range below the reflection boundary identified by the identification unit 13A, or may correct the recognition frame by masking a range below the reflection boundary identified by the identification unit 13A. When masking the recognition frame, the correction unit 13 further includes a mask processing unit 13B. The mask processing unit 13B corrects the recognition frame by applying mask processing to the range below the reflection boundary identified by the identification unit 13A. The method of correcting the range of the object by the correction unit 13 will be described later.

[0020] The output control unit 14 generates object information about objects recognized by the recognition unit 12 and outputs it to the output device 30. Specifically, the output control unit 14 causes the output device 30 to transmit the object information to another device. The object information output by the output control unit 14 also includes object information about objects whose recognition frames have been corrected by the correction unit 13. The object information output by the output control unit 14 may include, for example, information about the type of recognized object, and the object's position, distance, movement direction, movement speed, etc., calculated using a known method based on the recognition frame of the recognized object. The object information may include, for example, whether or not the object was recognized by the recognition unit 12 and the number of recognized objects. If the range of the object has been corrected, the output control unit 14 generates object information based on the corrected range of the object.

[0021] Here, if the ground on which the object exists is wet, a reflected image of the object reflected on the ground may appear in the image taken by the camera 21. In this case, the recognition unit 12 may erroneously recognize the reflected image of the object as part of the object, and may not be able to accurately obtain the position and distance of the recognized object.

[0022] FIG. 2 schematically illustrates an example of an image F1 including an object O1. FIG. 2 illustrates an example of an excerpt of the periphery of an object recognized by the recognition unit 12 in an image acquired by the image acquisition unit 11. In FIG. 2, the x direction is the horizontal or left-right direction of the image frame, and the y direction is the vertical or up-down direction of the image frame. In FIG. 2, due to the reflection image R1 of a person object O1 being reflected in a puddle P1 on the road, the recognition frame DE1 in image F1 extends downward beyond the actual range of object O1. In other words, the recognition unit 12 erroneously recognizes the range including the reflection image R1 as object O1 and assigns the recognition frame DE1 to the range of object O1. As a result, the recognition frame DE1 includes a lower area than the actual recognition frame of object O1. The object distance included in the object information is calculated based on the coordinates in the image of the bottom edge of the recognition frame, which is the ground position of the object. The distance from the camera 21 to the object or the distance from the multi-functional utility pole to the object is calculated. Therefore, as shown in Figure 2, if a recognition frame is assigned to the range including the reflected image R1, object information will be generated that indicates that the object is located closer to the camera 21 or the multi-functional utility pole than the actual distance to the object.

[0023] FIG. 3 illustrates an example of brightness values ​​in the vertical direction within the recognition frame of FIG. 2. FIG. 3 illustrates, for example, brightness values ​​in the vertical direction at the center in the horizontal direction within the recognition frame of FIG. 2. In FIG. 3, the vertical axis indicates the brightness values ​​of pixels in the vertical direction within the recognition frame, and the horizontal axis indicates the position of the pixel in the vertical direction of the recognition frame in the image. In other words, the graph showing brightness values ​​in FIG. 3 shows brightness values ​​from the top to bottom at the center in the horizontal direction of the recognition frame DE1 in FIG. 2. Pixels to the right of the pixel position indicated by X3 in FIG. 3 are pixels in the range where the recognition frame DE1 shown in FIG. 2 overlaps with the puddle P1. As shown in FIG. 3, in the range corresponding to the puddle P1, far-infrared rays emitted by the object O1 are reflected by the puddle P1, are greatly attenuated, and then enter the camera 21. In other words, brightness tends to be greatly attenuated near the reflection boundary between the recognition target range in the recognition frame DE1 where the object exists and the reflection range where the puddle P1 exists. The recognition processing device 10 of this embodiment utilizes the attenuation characteristics of light due to this reflection to correct the recognition frame of the object. A method for correcting the range of an object according to the first embodiment will now be described.

[0024] First, the identification unit 13A of the first embodiment applies a low-pass filter to the luminance of the pixel located at the center in the horizontal direction of the recognition frame as preprocessing, and then calculates the gradient value of the luminance after the application. Fig. 4 illustrates an example of the gradient value of the luminance within the recognition frame of Fig. 2. In Fig. 4, the vertical axis represents the gradient value of the luminance of the pixel in the vertical direction of the recognition frame, and the horizontal axis represents the position of the pixel in the up-down direction of the recognition frame in the image. As described above, light incident on puddle P1 is significantly attenuated by puddle P1, and therefore, as shown in Fig. 4, the gradient value fluctuates significantly near the boundary of puddle P1.

[0025] Next, the determination unit 13A determines whether the maximum amplitude of the calculated luminance gradient value is equal to or greater than a predetermined amplitude threshold. The maximum amplitude here is the maximum peak-to-peak value (peak-to-peak value) that represents the difference between the peak value and the adjacent peak value in the calculated luminance gradient value. In the example of FIG. 4, the determination unit 13A determines the difference between the maximum gradient value Gmax at pixel position X1 and the minimum gradient value Gmin at pixel position X2 as the maximum amplitude Am, and determines whether the maximum amplitude Am is equal to or greater than a predetermined amplitude threshold. In the example of FIG. 4, the maximum amplitude Am is assumed to be equal to or greater than the predetermined amplitude threshold.

[0026] Next, the identification unit 13A identifies a reflection boundary between the recognition target range and the reflection range when the maximum amplitude is equal to or greater than a predetermined amplitude threshold. For example, the identification unit 13A identifies the reflection boundary by determining, as a reference position Lr in the vertical direction of the reflection boundary, the position X3 that is the uppermost position below the position corresponding to the minimum value Gmin of the gradient value and where the change in the gradient value (i.e., the second-order differential value of luminance) is equal to or less than a predetermined gradient change threshold. In this embodiment, the identification unit 13A identifies the reflection boundary using a straight line y=Lr based on the reference position Lr. Here, it has been experimentally found that the position where the gradient value waveform begins to stabilize below the range where the maximum amplitude occurs approximately coincides with the position of the reflection boundary. Therefore, it is preferable to set the gradient change threshold to a value close to 0, for example. In the example of FIG. 4, the pixel position corresponding to pixel position 156 in the vertical direction is the reference position Lr and is the reflection boundary.

[0027] Fig. 5 is a diagram illustrating an example of a reflection boundary B in image F1. In the example of Fig. 5, the reflection boundary B is shown between a recognition target range A1 in which object O1 exists and a reflection range A2 in which reflected image R1 exists within recognition frame DE1.

[0028] Next, the modification unit 13 modifies the recognition frame of the object O1 from the recognition frame DE1 to a recognition frame whose lower end is the reflection boundary B, i.e., a recognition frame consisting of the recognition target range A1, to generate a recognition frame DE2. If the modification unit 13 includes a mask processing unit 13B, the mask processing unit 13B applies mask processing to the range below the reflection boundary B, i.e., the reflection range A2, to modify the recognition frame DE1 so as to exclude the reflection range A2 from the recognition frame DE1, thereby generating a modified recognition frame DE2.

[0029] FIG. 6 is a diagram schematically illustrating an example of an image F1 after the recognition frame has been corrected. The correction unit 13 of the first embodiment generates a corrected recognition frame DE2 that has been corrected so as not to include the reflection range A2 below the reflection boundary B identified in the recognition frame. When masking the reflection range A2, the mask processing unit 13B performs masking so as to cover it with a mask M1. The correction unit 13 generates a corrected recognition frame DE2 so that the boundary of the recognition frame DE1 on the mask M1 side is positioned so as to contact the upper end of the mask M1. This corrects the range of the object, and the appropriate ground position of the object can be obtained.

[0030] 7 is a flowchart showing an example of a recognition processing method according to the first embodiment. The processing shown in FIG.

[0031] In step S101, the image acquisition unit 11 acquires an image captured by the camera 21.

[0032] In step S102, the recognition unit 12 performs object recognition processing on the image acquired in step S101 and determines whether or not an object has been recognized in the image. If it is determined that an object has been recognized (Yes in step S102), the recognition unit 12 generates a recognition frame indicating the range of the recognized object and proceeds to step S103. If it is not determined that an object has been recognized (No in step S102), the processing shown in FIG. 7 ends.

[0033] In step S103, the specification unit 13A calculates the gradient value of brightness in the vertical direction in the recognition frame of the object.

[0034] In step S104, the identification unit 13A determines whether the maximum amplitude of the calculated brightness gradient value is equal to or greater than a predetermined amplitude threshold. If it is determined that the maximum amplitude is equal to or greater than the predetermined threshold (Yes in step S104), the identification unit 13A determines that the reflection range is within the recognition frame, and proceeds to step S105. If it is determined that the maximum amplitude is not equal to or greater than the predetermined threshold (No in step S104), the processing shown in FIG. 7 ends.

[0035] In step S105, the identification unit 13A identifies a reflection boundary between the recognition target range and a reflection range in which a non-recognition target object different from the recognition target object exists.

[0036] In step S106, the modifying unit 13 modifies the recognition frame based on the reflecting boundary. Specifically, the modifying unit 13 modifies the recognition frame so as not to include the range below the reflecting boundary. After step S107, the processing shown in FIG. 7 ends.

[0037] In the process of step S106, the mask processing unit 13B may perform mask processing on the reflection range below the reflection boundary to correct the recognition frame.

[0038] In the first embodiment, the recognition processing device 10 includes an image acquisition unit 11 that acquires an image, a recognition unit 12 that recognizes an object from the image, and a correction unit 13 that corrects the range of the recognized object based on the manner in which brightness attenuates in the vertical direction of the image frame within the range of the recognized object. According to this configuration, correcting the range of the object based on the manner in which brightness attenuates in the image can suppress erroneous recognition due to the reflection of the object in a body of water on the ground surface. Therefore, objects can be recognized more accurately in the image recognition process, and accurate object information can be generated.

[0039] In the recognition processing device 10 of the first embodiment, based on the attenuation state in which the maximum amplitude of the luminance gradient value is equal to or greater than a predetermined amplitude threshold, the modifying unit 13 modifies the range of the object so as not to include a range below a position corresponding to the minimum value of the gradient value and below a position where the change in the gradient value is equal to or less than a predetermined gradient change amount threshold. With this configuration, it is possible to assign a recognition frame of an appropriate range within the range of the object that does not include a range in which a reflected image of the object is reflected.

[0040] Furthermore, as a modification of the first embodiment, the recognition processing device 10 may modify the object range based on the attenuation pattern in which the brightness decreases below the range of the object, so that the modifying unit 13 does not include the range where the brightness is decreased. As shown in Fig. 3, the brightness value in the vertical direction within the recognition frame of the object decreases downward from the position of the pixel indicated by X3 as the boundary. The modifying unit 13 modifies the recognition frame so as not to include the range below the position where the brightness is decreased as shown in Fig. 3.

[0041] Second embodiment 8 is a block diagram schematically showing the functional configuration of a recognition device 1A according to the second embodiment. The recognition device 1A according to the second embodiment differs from the recognition device 1 according to the first embodiment in that the functional block of the recognition processing device 10A includes an object determination unit 15 and an estimation unit 16. The following description of the second embodiment will focus on the differences from the first embodiment, and will omit a description of the commonalities as appropriate.

[0042] The object determination unit 15 determines whether an object recognized by the recognition unit 12 is a person. If the object recognized by the recognition unit 12 is an object recognized by a person recognition model, the object determination unit 15 determines that the recognized object is a person. The recognition unit 12 may assign a flag indicating the type of object to the recognized object, and the object determination unit 15 may determine that the recognized object is a person based on the flag assigned by the recognition unit 12.

[0043] The estimation unit 16 estimates or identifies the lower body part of a recognized object (person). For example, the estimation unit 16 estimates or identifies the lower body part of a recognized person based on the person recognition result by the recognition unit 12. Specifically, the recognition unit 12 recognizes a person using a whole-body model obtained by machine learning a whole-body image of the person and a partial-body model obtained by machine learning an image of the person's upper body, lower body, etc. Furthermore, when the person recognized by the recognition unit 12 using the whole-body model overlaps with the lower body of a person recognized using the partial-body model, the estimation unit 16 identifies the lower body part of the person recognized using the whole-body model as the lower body part recognized in the overlapping area. Furthermore, the estimation unit 16 may estimate the lower body part by performing a skeleton estimation process on the recognized person's image. Specifically, the estimation unit 16 estimates the positions of the object's joints based on skeleton data output by inputting the image of the person recognized by the recognition unit 12 into a skeleton estimation model. This skeleton estimation model is a trained model that has been machine-learned using a known technique to input a person's image and output the person's skeleton data. The estimation unit 16 estimates the lower body part of the person from the arrangement of the person's joints estimated by the skeleton estimation model.

[0044] A method for correcting the range of an object according to the second embodiment will be described with reference to Fig. 9. Fig. 9 illustrates an example of a state in which a skeleton estimation process has been performed on an image F1. As shown in Fig. 9, a skeleton S estimated by the estimation unit 16 is shown within an object O1, which is a person. The skeleton S also shows a neck joint J1, shoulder joints J2 and J3, elbow joints J4 and J5, waist joint J6, and knee joints J7 and J8 estimated by the estimation unit 16. The identification unit 13A of the second embodiment calculates a brightness gradient value in a range below the estimated position of the waist joint J6.

[0045] Fig. 10 is a flowchart showing an example of a recognition processing method according to the second embodiment. Steps S201 to S202, S204, and S207 to S209 in Fig. 10 are the same as steps S101 to S106 in Fig. 7, and therefore description thereof will be omitted.

[0046] In step S203, the object determination unit 15 determines whether the recognized object is a person. If the object is a person (Yes in step S203), the process proceeds to step S205. If the object is not a person (No in step S203), the process proceeds to step S204, and in steps S204, S207 to S210, the recognition frame is corrected as necessary.

[0047] In step S205, the estimation unit 16 estimates the position of the waist joint of the recognized human object.

[0048] In step S206, the identification unit 13A calculates the gradient value of brightness in the up-down direction in the range below the position of the waist joint J6 in the recognition frame of the object. Then, the process proceeds to step S207, and in steps S207 to S210, the recognition frame is corrected as necessary, and the process shown in FIG. 10 ends.

[0049] In the second embodiment, the gradient value is calculated for the range below the waist joint J6 in the recognition frame, but this is not limiting. The gradient value may be calculated for the range below any position in the lower body of the person, such as the waist joint J6, the knee joints J7 and J8, or the portion between the waist joint J6 and the knee joint J7 or J8.

[0050] The recognition processing device 10 of the second embodiment includes an object determination unit 15 that determines whether a recognized object is a person, and an estimation unit 16 that estimates the joints of the person's lower body if the recognized object is a person. A correction unit 13 corrects the range of the object based on the attenuation pattern in the range below the estimated position of the joints of the lower body within the range of the object. Here, since the puddle exists at the contact point between the object and the road surface, it is located lower in the recognition frame. This configuration makes it possible to appropriately remove the reflection range from the recognition frame while reducing the processing load for calculating the gradient value.

[0051] Several variations of the present disclosure will be described below.

[0052] In the above embodiment, the recognition frame is corrected based on the luminance of the pixel located at the center in the horizontal direction of the recognition frame, but this is not limiting. For example, the recognition frame may be corrected based on the luminance of a pixel located at a position shifted to the right or left from the center in the horizontal direction of the recognition frame, or based on the average luminance of multiple pixels in the horizontal direction of the recognition frame.

[0053] In the above embodiment, the recognition frame is corrected when the maximum amplitude of the brightness gradient value is equal to or greater than the amplitude threshold. However, this is not limiting. For example, the recognition frame may be corrected when the brightness attenuates from a value greater than a predetermined brightness threshold and then continuously falls to a value equal to or less than the predetermined brightness threshold. In this case, the reflection boundary B may be identified by any position within the range where the brightness is continuously equal to or less than the predetermined brightness threshold.

[0054] The present disclosure has been described above with reference to the above-mentioned embodiments, but the present disclosure is not limited to the above-mentioned embodiments, and appropriate combinations or substitutions of the configurations shown in the embodiments are also included in the present disclosure. [Explanation of symbols]

[0055] 1, 1A...recognition device, 10, 10A...recognition processing device, 11...image acquisition unit, 12...recognition unit, 13...correction unit, 13A...identification unit, 13B...mask processing unit, 14...output control unit, 15...object determination unit, 16...estimation unit, 21...camera, 30...output device

Claims

1. an image acquisition unit that acquires an image; a recognition unit that recognizes an object from the image; a correction unit that corrects the range of the object based on a luminance attenuation pattern in the vertical direction of the image frame within the range of the recognized object; A recognition processing device comprising:

2. The recognition processing device according to claim 1 , wherein the correction unit corrects the range of the object so as not to include the range where the brightness is reduced, based on the attenuation mode in which the brightness is reduced below the range of the object.

3. 2. The recognition processing device according to claim 1, wherein the modification unit modifies the range of the object based on the attenuation state in which the maximum amplitude of the luminance gradient value is equal to or greater than a predetermined amplitude threshold, so as not to include a range below a position corresponding to a minimum value of the gradient value and below a position where an amount of change in the gradient value is equal to or less than a predetermined gradient change amount threshold.

4. an object determination unit that determines whether the object recognized by the recognition unit is a person; an estimation unit that estimates a lower body part of the person recognized by the recognition unit; Equipped with The recognition processing device according to claim 1 , wherein the correction unit corrects the range of the object based on the attenuation state in a range below the estimated position of the lower body part within the range of the object relative to the recognized person.

5. acquiring an image; Recognizing an object from the image; modifying the range of the recognized object based on the manner in which brightness attenuates in the vertical direction of the image frame within the range of the recognized object; A recognition processing method comprising:

6. On the computer, acquiring an image; Recognizing an object from the image; modifying the range of the recognized object based on the manner in which brightness attenuates in the vertical direction of the image frame within the range of the recognized object; A recognition processing program that executes the above.

Images