Pedestrian re-identification method and device based on RGBD image and storage medium

Abstract

The application discloses a pedestrian re-identification method and device based on an RGBD image and a storage medium, and the method comprises the following steps: acquiring an RGB image and a depth image which comprise the same target scene and are aligned, and performing foreground segmentation on the depth image to obtain a depth foreground image; determining an RGB foreground image by using the depth foreground image according to the alignment relationship between the RGB image and the depth image, and performing pedestrian target detection on the RGB foreground image to determine a first pedestrian target image corresponding to the RGB image and a second pedestrian target image corresponding to the depth image; determining a single-person RGB image according to the first pedestrian target image and the second pedestrian target image, and performing person feature extraction and matching on the single-person RGB image to obtain a pedestrian re-identification result. The application increases the depth image on the basis of the RGB image to perform the pedestrian re-identification, solves the background interference, facilitates the identification of specific pedestrians in cross-view images, and improves the accuracy of the pedestrian re-identification.

Description

Technical Field

[0001] This invention relates to the field of pedestrian re-identification technology, and in particular to a pedestrian re-identification method, apparatus, terminal device and storage medium based on RGBD images. Background Technology

[0002] Person re-identification, also known as ReID, is a method that uses computer vision algorithms to find a specific person in a sequence of images or videos across different fields of view. However, due to factors such as camera resolution and shooting angle, it is often impossible to obtain clear facial images, affecting the accuracy of person re-identification.

[0003] Currently, the main methods used for person re-identification are deep learning-based, including: ReID methods based on representation learning, ReID methods based on metric learning, ReID methods based on local features, ReID methods based on video sequences, and ReID methods based on GAN image generation. However, these methods all utilize only the RGB information in RGB images, which contain too much background interference, hindering person re-identification and resulting in low accuracy.

[0004] Therefore, existing technologies still need to be improved and enhanced. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a pedestrian re-identification method, device, terminal equipment and storage medium based on RGBD images, in order to address the above-mentioned defects of the prior art. The aim is to solve the problem that the existing pedestrian re-identification methods all use pure RGB images, and the recognition accuracy is low due to the large amount of background interference in RGB images.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0007] In a first aspect, the present invention provides a pedestrian re-identification method based on RGBD images, wherein the method includes: acquiring RGB images and depth images that include the same target scene and are aligned, and performing foreground segmentation on the depth image to obtain a depth foreground image corresponding to the depth image; determining the RGB foreground image corresponding to the RGB image using the depth foreground image based on the alignment relationship between the RGB image and the depth image, and performing pedestrian target detection on the RGB foreground image to determine a first pedestrian target image corresponding to the RGB image and a second pedestrian target image corresponding to the depth image; determining a single-person RGB image corresponding to the RGB image based on the first pedestrian target image and the second pedestrian target image, and performing human feature extraction and matching on the single-person RGB image to obtain a pedestrian re-identification result.

[0008] Secondly, the present invention provides a pedestrian re-identification device based on RGBD images, wherein the device includes: a depth foreground image determination module, used to acquire RGB images and depth images that include the same target scene and are aligned, and to perform foreground segmentation on the depth image to obtain a depth foreground image corresponding to the depth image; a pedestrian target image determination module, used to determine the RGB foreground image corresponding to the RGB image based on the alignment relationship between the RGB image and the depth image and the depth foreground image, and to perform pedestrian target detection on the RGB foreground image to determine a first pedestrian target image corresponding to the RGB image and a second pedestrian target image corresponding to the depth image; and a pedestrian re-identification module, used to determine a single-person RGB image corresponding to the RGB image based on the first pedestrian target image and the second pedestrian target image, and to perform human feature extraction and matching on the single-person RGB image to obtain a pedestrian re-identification result.

[0009] Thirdly, embodiments of the present invention also provide a terminal device, wherein the terminal device includes a memory, a processor, and a program for a pedestrian re-identification method based on RGBD images stored in the memory and executable on the processor. When the processor executes the program for the pedestrian re-identification method based on RGBD images, it implements the steps of the pedestrian re-identification method based on RGBD images in any of the above-described schemes.

[0010] Fourthly, embodiments of the present invention also provide a computer-readable storage medium, wherein the computer-readable storage medium stores a program for a pedestrian re-identification method based on RGBD images, and when the program for the pedestrian re-identification method based on RGBD images is executed by a processor, it implements the steps of the pedestrian re-identification method based on RGBD images in any of the above-described schemes.

[0011] Beneficial effects: Compared with the prior art, the present invention provides a pedestrian re-identification method based on RGBD images. On the basis of RGB images, a depth image is added for pedestrian re-identification, which solves the background interference, makes it easier to identify specific pedestrians in images with cross-field of view, and improves the accuracy of pedestrian re-identification. Attached Figure Description

[0012] Figure 1 A flowchart illustrating a specific implementation of the pedestrian re-identification method based on RGBD images provided in this embodiment of the invention.

[0013] Figure 2 This is a schematic diagram of the pedestrian re-identification device based on RGBD images provided in an embodiment of the present invention.

[0014] Figure 3 This is a block diagram illustrating the internal structure of the terminal device provided in an embodiment of the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and effects of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0016] Figure 1 This is a flowchart illustrating a pedestrian re-identification method based on RGBD images according to an embodiment of the present invention. The method described in this embodiment can improve the accuracy of pedestrian re-identification to a certain extent. This method can be applied to terminal devices, such as computers, mobile phones, smart TVs, and other intelligent products. In specific applications, the terminal device can have image acquisition capabilities or be connected to a preset image acquisition device to acquire RGB images and depth images. Specifically, as... Figure 1 As shown in the figure, the pedestrian re-identification method based on RGBD images in this embodiment includes the following steps:

[0017] Step S100: Obtain RGB images and depth images that include the same target scene and are aligned, and perform foreground segmentation on the depth image to obtain the depth foreground image corresponding to the depth image.

[0018] In this embodiment, the RGBD image consists of two images: an RGB three-channel color image and a depth image. The depth image is essentially a grayscale image, where each pixel value represents the actual distance between the sensor and the object. Therefore, this embodiment first needs to acquire both the RGB image and the depth image. Since the depth image is unaffected by lighting or shadows, and each pixel value in the depth image reflects the actual distance between the sensor and the object, this embodiment can determine the corresponding depth foreground image based on the depth image, which is beneficial for identifying the target pedestrian in subsequent steps.

[0019] In specific implementation, this embodiment can acquire RGB images and depth images through a preset image acquisition device, which can be a camera installed on the terminal device. After the camera is activated, it captures images of the target scene, thereby acquiring RGB images and depth images. In this embodiment, the target scene can be a designated intersection, a designated street, or a designated road segment. When acquiring RGB images and depth images, this embodiment saves the RGB images and depth images as a sequence of frames. While acquiring images of the target scene, this embodiment can also acquire a small number of background images of the same target scene, for example, acquiring 50 background images without pedestrians, and storing these background images before all the sequence frames.

[0020] In one embodiment, since the RGB image and the depth image are acquired based on the same target scene, after the RGB image and the depth image are acquired, they need to be aligned. The purpose of alignment is to align the pixels of the two images so that there is a one-to-one correspondence between the pixels, so that the background area and the foreground area of ​​the RGB image can be determined by the background modeling mapping of the depth image in subsequent steps.

[0021] In one embodiment, each pixel value in the acquired depth image reflects the actual distance between the sensor and the object, i.e., it embodies the depth information of each pixel. This depth information is not affected by light, shadows, etc. Therefore, background modeling based on the depth image can effectively segment the foreground region in the depth image, thus obtaining a depth foreground image. For example, when the depth image includes both people and snow, the background information containing only snow is first learned. When people appear, the foreground image containing only people and not snow can be segmented through background modeling.

[0022] It should be noted that the background modeling methods in this embodiment include: frame difference method, Gaussian mixture background modeling, optical flow method, etc., and no limitation is made here.

[0023] Step S200: Based on the alignment relationship between the RGB image and the depth image, determine the RGB foreground image corresponding to the RGB image using the depth foreground image, and perform pedestrian target detection on the RGB foreground image to determine the first pedestrian target image corresponding to the RGB image and the second pedestrian target image corresponding to the depth image.

[0024] In one embodiment, based on the alignment relationship between the RGB image and the depth image, after obtaining the depth foreground image, the corresponding RGB foreground image can be determined using this depth foreground image, i.e., the foreground region in the RGB image is obtained. Since the RGB foreground image reflects the foreground region, after obtaining the RGB foreground image, this embodiment can perform pedestrian target detection on the RGB foreground image, i.e., detect pedestrians from the foreground region, thereby obtaining a first pedestrian target image corresponding to the RGB image. This first pedestrian target image reflects that there are only pedestrians in the RGB image, and no other targets. Further, after obtaining the first pedestrian target image corresponding to the RGB image, based on the alignment relationship between the depth image and the RGB image and the first pedestrian target image corresponding to the RGB image, a second pedestrian target image corresponding to the depth image is determined from the depth foreground image. This second pedestrian target image reflects that there are only pedestrians in the depth image, and no other targets.

[0025] In one implementation, the determination of the RGB foreground image in this embodiment includes the following steps:

[0026] Step S201: Denoise the depth foreground image to obtain the denoised depth foreground image.

[0027] It should be noted that, in order to obtain a more accurate RGB foreground image, this embodiment performs denoising processing on the depth foreground image after obtaining it, resulting in a denoised depth foreground image. The denoising processing in this embodiment includes morphological operations and binarization operations. Morphological operations include image dilation and erosion, opening and closing operations, etc., which are not limited here.

[0028] Step S202: Obtain the pixel information corresponding to the depth foreground image after denoising.

[0029] Step S203: Based on the alignment relationship between the RGB image and the depth image, map the pixel information to the RGB image to obtain the RGB foreground image.

[0030] Since the RGB image and the depth image are aligned and there is a one-to-one correspondence between their pixels, this embodiment obtains the pixel information corresponding to the denoised depth foreground image after obtaining the depth foreground image. Then, it maps this pixel information onto the RGB image to determine the region corresponding to the depth foreground image in the RGB image. This region is the foreground region in the RGB image, thus obtaining the RGB foreground image. This embodiment determines the RGB foreground image by mapping the depth foreground image onto the RGB image. Compared to existing technologies that directly use RGB images for background modeling, which are easily affected by lighting, shadows, etc., this embodiment facilitates accurate extraction of the foreground image from the RGB image, removing background interference from the RGB image.

[0031] In one implementation, this embodiment includes the following steps when determining the first pedestrian target map and the second pedestrian target map:

[0032] Step S21: Perform pedestrian target detection on the RGB foreground image to determine the pedestrians in the RGB foreground image and their corresponding location information.

[0033] Specifically, the RGB foreground image reflects the foreground region in an RGB image. This foreground region can include various foreground categories, such as pedestrians and vehicles. Since the substantive information reflected by different foreground categories is different, in one embodiment, by identifying various foreground categories in the RGB foreground image and assigning corresponding label information to each category, and then filtering out the foreground labeled "pedestrian" based on the label information, the pedestrian target in the RGB foreground image can be obtained, thereby achieving pedestrian target detection in the RGB foreground image. Furthermore, after filtering pedestrians from the RGB foreground image, the location information of the pedestrians can also be determined. This location information reflects the coordinate information of the pedestrians in the RGB foreground image, thereby determining the position of each pedestrian in the image.

[0034] Step S22: Based on the pedestrian and location information and the alignment relationship between the RGB image and the depth image, crop the pedestrian region in the RGB image and the depth image respectively to obtain the first pedestrian target and the second pedestrian target image.

[0035] In one embodiment, based on pedestrians and their corresponding location information, a first pedestrian target image is determined from the RGB foreground image. This first pedestrian target image is obtained by removing all content except pedestrians from the RGB foreground image; that is, the first pedestrian target image contains only pedestrians. Since the RGB image and the depth image are aligned, the RGB foreground image and the depth foreground image are also aligned. Therefore, the first pedestrian target image can be mapped onto the depth foreground image to obtain a second pedestrian target image. This second pedestrian target image is also obtained by removing all content except pedestrians from the depth foreground image; that is, the second pedestrian target image contains only pedestrians.

[0036] Step S300: Determine the single-person RGB image corresponding to the RGB image based on the first pedestrian target image and the second pedestrian target image, and extract and match the human features of the single-person RGB image to obtain the pedestrian re-identification result.

[0037] Since the first pedestrian target image in this embodiment is determined from the RGB foreground image, it can reflect the pedestrians and their position information in the RGB image; the second pedestrian target image is determined from the depth foreground image, thus reflecting the pedestrians and their position information in the depth image. This embodiment identifies pedestrians to determine the target pedestrian. Therefore, based on the first and second pedestrian target images, this embodiment determines the single-person RGB image corresponding to the RGB image. This single-person RGB image is a color image containing only one pedestrian. After obtaining the single-person RGB image, this embodiment can extract and match the features of each single-person RGB image to determine the target pedestrian and complete pedestrian re-identification.

[0038] In one implementation, this embodiment includes the following steps when determining a single person's RGB image:

[0039] Step S301: Perform pedestrian segmentation on the second pedestrian target image to obtain the single-person depth map corresponding to the second pedestrian target image;

[0040] Step S302: Based on the alignment relationship between the color image and the depth image, map the single person depth image to the first pedestrian target image to obtain the single person RGB image.

[0041] In practical implementation, since the second pedestrian target image reflects the pedestrians and their position information in the depth image, it is also a grayscale image. This embodiment solves the pedestrian occlusion problem by segmenting the pedestrians in the second pedestrian target image to obtain several individual depth images. Then, based on the alignment relationship between the color image and the depth image, the segmented individual depth images are mapped onto the first pedestrian target image. This allows the position of each individual depth image to be determined in the first pedestrian target image, effectively segmenting the pedestrians in the first pedestrian target image to obtain individual RGB images. Since the individual RGB image in this embodiment only includes one pedestrian, background interference and irrelevant events are removed, making it more conducive to pedestrian re-identification.

[0042] In one embodiment, segmenting pedestrians in the second pedestrian target image to obtain several individual depth maps more specifically includes: dividing the pixels of the second pedestrian target image into several pixel groups and randomly initializing the center point of each pixel group; calculating the distance from each pixel to the center point, classifying the pixel group according to the distance between each pixel and the center point, that is, classifying the pixel group to the pixel group corresponding to the nearest center point; traversing all pixels, recalculating the center point of each pixel group according to the pixel group classification, and repeating the above steps until the coordinates of the center of each pixel group no longer change or the change of the center coordinates is within the error range. Thus, each pixel group is a pedestrian, realizing the segmentation of pedestrians in the second pedestrian target image to obtain several individual depth maps.

[0043] In one implementation, this embodiment includes the following steps when performing pedestrian re-identification:

[0044] Step S31: Use the preset ReID model to extract the features of a single person's RGB image to obtain the feature map corresponding to each frame of the single person's RGB image;

[0045] Step S32: Match each frame feature map with a preset standard map to determine the similarity between each frame feature map and the preset standard map;

[0046] Step S33: Based on the similarity, determine whether the pedestrian in the RGBD image is the target pedestrian to obtain the pedestrian re-identification result.

[0047] Specifically, in this embodiment, when recognizing each frame of a single-person RGB image, a ReID model can be used to extract facial features, thus obtaining a feature map corresponding to each frame of the single-person RGB image. This feature map reflects the facial features of each pedestrian. After obtaining the feature map, this embodiment matches each frame's feature map with a preset standard image. The preset standard image is a pre-stored facial feature map of the target pedestrian, which can be the facial feature map corresponding to the target pedestrian captured by different cameras in different fields of view or by a single camera at different time periods, to achieve cross-field-of-view pedestrian re-identification. Therefore, based on matching each frame's feature map with the preset standard image, the similarity between each frame's feature map and the preset standard image can be determined. This similarity reflects the degree of similarity between the feature map extracted from each frame of the single-person RGB image and the preset standard image.

[0048] Furthermore, to accurately identify the target pedestrian, this embodiment compares the obtained similarity score with a preset similarity threshold. This similarity threshold is used to measure and determine whether the feature map and the preset standard map belong to the same pedestrian. Therefore, when the similarity of a feature map is greater than or equal to the similarity threshold, the pedestrian in the single-person RGB image corresponding to that feature map can be identified as the target pedestrian, thereby achieving cross-view pedestrian re-identification.

[0049] In summary, this embodiment adds a depth image to the RGB image for pedestrian re-identification, which solves the problem of background interference, makes it easier to identify specific pedestrians in images across different fields of view, and improves the accuracy of pedestrian re-identification.

[0050] Figure 2 This invention provides a schematic diagram of a pedestrian re-identification device based on RGBD images. The pedestrian re-identification device in this embodiment includes: a depth foreground image determination module 10, a pedestrian target image determination module 20, and a pedestrian re-identification module 30. Specifically, the depth foreground image determination module 10 is used to acquire RGB images and depth images that include the same target scene and are aligned, and to perform foreground segmentation on the depth image to obtain a depth foreground image corresponding to the depth image. The pedestrian target image determination module 20 is used to determine the RGB foreground image corresponding to the RGB image based on the alignment relationship between the RGB image and the depth image, using the depth foreground image, and to perform pedestrian target detection on the RGB foreground image to determine a first pedestrian target image corresponding to the RGB image and a second pedestrian target image corresponding to the depth image. The pedestrian re-identification module 30 is used to determine a single-person RGB image corresponding to the RGB image based on the first and second pedestrian target images, and to extract and match human features from the single-person RGB image to obtain the pedestrian re-identification result.

[0051] In one implementation, the pedestrian target image determination module 20 includes: a denoising processing unit, used to denoise the depth foreground image to obtain a denoised depth foreground image, the denoising processing including: morphological operations and binarization operations; a pixel information acquisition unit, used to acquire the pixel information corresponding to the denoised depth foreground image; and an RGB foreground image acquisition unit, used to map the pixel information to the RGB image according to the alignment relationship between the RGB image and the depth image to obtain the RGB foreground image.

[0052] In one implementation, the pedestrian target image determination module 20 further includes: a pedestrian detection unit, used to perform pedestrian target detection on the RGB foreground image and determine the pedestrians in the RGB foreground image and the corresponding position information of the pedestrians; and a target image determination unit, used to crop the pedestrian regions in the RGB image and the depth image respectively to obtain the first pedestrian target and the second pedestrian target image based on the pedestrians and position information and the alignment relationship between the RGB image and the depth image.

[0053] In one implementation, the pedestrian re-identification module 30 includes: a single-person depth map determination unit, used to perform pedestrian segmentation on the second pedestrian target map to obtain a single-person depth map corresponding to the second pedestrian target; and a single-person RGB image determination unit, used to map the single-person depth map to the first pedestrian target map according to the alignment relationship between the RGB image and the depth image to obtain a single-person RGB image.

[0054] In one implementation, the pedestrian re-identification module 30 further includes: a feature map determination unit, used to extract human features from a single-person RGB image using a preset ReID model to obtain a feature map corresponding to each frame of the single-person RGB image; wherein, the preset standard image is a pre-stored facial feature map of the target pedestrian; a similarity determination unit, used to match each frame of feature map with the preset standard image to determine the similarity between each frame of feature map and the preset standard image; and a target pedestrian determination unit, used to determine whether the pedestrian in the RGBD image is the target pedestrian based on the similarity to obtain the pedestrian re-identification result.

[0055] The working principle of the pedestrian re-identification device based on RGBD images in this embodiment is the same as that described in the above method embodiments, and will not be repeated here.

[0056] Based on the above embodiments, the present invention also provides a terminal device, the principle block diagram of which can be as follows: Figure 3 As shown, the terminal device may include one or more processors 100. Figure 3(Only one is shown in the image), memory 101, and a computer program 102 stored in memory 101 and executable on one or more processors 100, such as a program for pedestrian re-identification of RGBD images. When one or more processors 100 execute computer program 102, they can implement the various steps in the method embodiment for pedestrian re-identification of RGBD images. Alternatively, when one or more processors 100 execute computer program 102, they can implement the functions of various modules / units in the apparatus embodiment for pedestrian re-identification of RGBD images, which is not limited here.

[0057] In some embodiments, the electronic device further includes an acquisition device, which includes an RGB camera and a depth camera, for acquiring RGB images and depth images respectively and transmitting them to the processor 100 to obtain a foreground mask. It should be noted that the RGB camera and the depth camera can be set independently or as a single unit. The depth camera can be a structured light camera, a TOF camera, a binocular camera, or a multi-view camera, etc., and there is no limitation here.

[0058] In one embodiment, the processor 100 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0059] In one embodiment, memory 101 may be an internal storage unit of an electronic device, such as a hard drive or RAM. Memory 101 may also be an external storage device of the electronic device, such as a plug-in hard drive, smart media card (SMC), secure digital (SD) card, flash card, etc. Furthermore, memory 101 may include both internal and external storage units. Memory 101 is used to store computer programs and other programs and data required by the terminal device. Memory 101 can also be used to temporarily store data that has been output or will be output.

[0060] Those skilled in the art will understand that Figure 3The block diagram shown is merely a partial structural diagram related to the present invention and does not constitute a limitation on the terminal device to which the present invention is applied. The specific terminal device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0061] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the methods described above. Any references to memory, storage, operational databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual operating data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A pedestrian re-identification method based on RGBD images, characterized in that, The method includes: Acquire RGB images and depth images that are aligned and have the same target scene, and perform foreground segmentation on the depth image to obtain a depth foreground image corresponding to the depth image; Based on the alignment relationship between the RGB image and the depth image, the RGB foreground image corresponding to the RGB image is determined using the depth foreground image, and pedestrian target detection is performed on the RGB foreground image to determine the first pedestrian target image corresponding to the RGB image and the second pedestrian target image corresponding to the depth image; wherein, the first pedestrian target image and the second pedestrian target image contain only pedestrians; Divide the pixels of the second pedestrian target image into several pixel groups and randomly initialize the center point of each pixel group; Calculate the distance from each pixel to each center point, and classify the pixel's affiliation based on the distance between each pixel and each center point; The center point of each pixel group is recalculated based on the ownership of each pixel point, and the calculation is iterated until the coordinates of the center of each pixel group no longer change or change within the error range, so as to obtain the single person depth map corresponding to the second pedestrian target map. Based on the alignment relationship between the RGB image and the depth image, the single person depth map is mapped to the first pedestrian target map to obtain the single person RGB image. A preset ReID model is used to extract human features from the single-person RGB image to obtain the feature map corresponding to each frame of the single-person RGB image; Each frame of the feature map is matched with a preset standard map to determine the similarity between each frame of the feature map and the preset standard map; wherein, the preset standard map is a facial feature map of the target pedestrian captured by different cameras in different fields of view or by the same camera in different time periods and stored in advance; Based on the similarity, determine whether the pedestrian in the RGBD image is the target pedestrian to obtain the pedestrian re-identification result.

2. The pedestrian re-identification method based on RGBD images according to claim 1, characterized in that, The step of determining the RGB foreground image corresponding to the RGB image based on the alignment relationship between the RGB image and the depth image, using the depth foreground image, includes: The depth foreground image is denoised to obtain a denoised depth foreground image. Obtain the pixel information corresponding to the denoised depth foreground image; Based on the alignment relationship between the RGB image and the depth image, the pixel information is mapped onto the RGB image to obtain the RGB foreground image.

3. The pedestrian re-identification method based on RGBD images according to claim 1, characterized in that, The step of performing pedestrian target detection on the RGB foreground image to determine the first pedestrian target image corresponding to the RGB image and the second pedestrian target image corresponding to the depth image includes: Perform pedestrian target detection on the RGB foreground image to determine the pedestrians in the RGB foreground image and their corresponding location information; Based on the pedestrian and location information, and according to the alignment relationship between the RGB image and the depth image, the pedestrian regions in the RGB image and the depth image are cropped respectively to obtain the first pedestrian target image and the second pedestrian target image.

4. A pedestrian re-identification device based on RGBD images, characterized in that, The device includes: The depth foreground image determination module is used to acquire RGB images and depth images that include the same target scene and are aligned, and to perform foreground segmentation on the depth image to obtain the depth foreground image corresponding to the depth image; The pedestrian target image determination module is used to determine the RGB foreground image corresponding to the RGB image based on the alignment relationship between the RGB image and the depth image, using the depth foreground image, and to perform pedestrian target detection on the RGB foreground image to determine the first pedestrian target image corresponding to the RGB image and the second pedestrian target image corresponding to the depth image; wherein, the first pedestrian target image and the second pedestrian target image contain only pedestrians; The pedestrian re-identification module is used to divide the pixels of the second pedestrian target image into several pixel groups and randomly initialize the center point of each pixel group; Calculate the distance from each pixel to each center point, and classify the pixel's affiliation based on the distance between each pixel and each center point; The center point of each pixel group is recalculated based on the ownership of each pixel point, and the calculation is iterated until the coordinates of the center of each pixel group no longer change or change within the error range, so as to obtain the single person depth map corresponding to the second pedestrian target map. Based on the alignment relationship between the RGB image and the depth image, the single person depth map is mapped to the first pedestrian target map to obtain the single person RGB image. A preset ReID model is used to extract human features from the single-person RGB image to obtain the feature map corresponding to each frame of the single-person RGB image; Each frame of the feature map is matched with a preset standard map to determine the similarity between each frame of the feature map and the preset standard map; wherein, the preset standard map is a facial feature map of the target pedestrian captured by different cameras in different fields of view or by the same camera in different time periods and stored in advance; Based on the similarity, determine whether the pedestrian in the RGBD image is the target pedestrian to obtain the pedestrian re-identification result.

5. A terminal device, characterized in that, The terminal device includes a memory, a processor, and a pedestrian re-identification program based on RGBD images stored in the memory and executable on the processor. When the processor executes the pedestrian re-identification program based on RGBD images, it implements the steps of the pedestrian re-identification method based on RGBD images as described in any one of claims 1-3.

6. The terminal device according to claim 5, characterized in that, The terminal device further includes an image acquisition device for acquiring depth images and RGB images respectively and transmitting them to the processor to implement the steps of the pedestrian re-identification method based on RGBD images as described in any one of claims 1-3.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a pedestrian re-identification program based on RGBD images. When the pedestrian re-identification program based on RGBD images is executed by a processor, it implements the steps of the pedestrian re-identification method based on RGBD images as described in any one of claims 1-3.

Images