Pedestrian image preprocessing method and device based on pedestrian re-identification

Abstract

The present invention provides a pedestrian image preprocessing method and device based on pedestrian re-identification. The pedestrian image preprocessing method includes: applying a first feature extraction module of a neural network model based on a pedestrian image to be processed to determine image features corresponding to the pedestrian image; applying an interpolation and scaling algorithm based on the image features to determine features of the interpolated and scaled image; applying a second feature extraction module and a third feature extraction module of the neural network model based on the interpolated and scaled image features to determine a scaled image; and applying a fourth feature extraction module and a perspective transformation module of the neural network model based on the interpolated and scaled image features to determine a perspective matrix corresponding to the image features; and applying a perspective transformation method based on the scaled image and the perspective matrix to determine a processed pedestrian image. The pedestrian image preprocessing method of the present invention can reduce the style influence caused by image preprocessing.

Description

Technical Field

[0001] The present invention relates to the field of neural networks, and in particular to a pedestrian image preprocessing method and device based on pedestrian re-identification. Background Art

[0002] Person re-identification (Re-ID) is a hot topic in computer vision research. This technology can identify pedestrians based on information such as their clothing, body shape, and hairstyle, enabling cross-camera and cross-scene recognition and retrieval. However, the type of camera, the angle of capture, and the actual environment of the scene all directly affect the quality of the captured pedestrian images, making the quality of pedestrian images extremely important.

[0003] In the existing technology, different cameras have different imaging sensors. In addition, due to differences in parameter adjustment, imaging conditions, and imaging angles, pedestrian images obtained by the same camera are often prone to different degrees of perspective distortion and image distortion at different positions in the picture. After obtaining the pedestrian image through the camera, it is necessary to first accurately locate the image position of the pedestrian, and then intercept the pedestrian image for effective input processing, which also requires the integrity of the pedestrian information.

[0004] Therefore, how to obtain high-quality pedestrian images before pedestrian re-identification processing is an urgent problem to be solved. Summary of the Invention

[0005] In view of the above-mentioned shortcomings of the prior art, the purpose of the present invention is to provide a pedestrian image preprocessing method and device based on pedestrian re-identification, which is used to solve the problem in the prior art that high-quality pedestrian images cannot be obtained before pedestrian re-identification processing.

[0006] To achieve the above-mentioned objectives and other related objectives, the present invention provides a pedestrian image preprocessing method based on pedestrian re-identification, comprising the following steps: based on the pedestrian image to be processed, applying the first feature extraction module of a neural network model to determine the image features corresponding to the pedestrian image; based on the image features, applying the interpolation and scaling algorithm to determine the features of the interpolated and scaled image; based on the interpolated and scaled image features, applying the second feature extraction module and the third feature extraction module of the neural network model to determine the scaled image; and based on the interpolated and scaled image features, applying the fourth feature extraction module and the perspective transformation module of the neural network model to determine the perspective matrix corresponding to the image features; based on the scaled image and the perspective matrix, applying the perspective transformation method to determine the processed pedestrian image.

[0007] In one embodiment of the present invention, the input channel value of the processed image of the first feature extraction module is less than the output channel value of the processed image of the first feature extraction module; the input channel value of the processed image of the second feature extraction module is equal to the output channel value of the processed image of the second feature extraction module; the input channel value of the processed image of the third feature extraction module is greater than the output channel value of the processed image of the third feature extraction module; the input channel value of the processed image of the fourth feature extraction module is equal to the output channel value of the processed image of the fourth feature extraction module.

[0008] In one embodiment of the present invention, based on the interpolated and scaled image features, the fourth feature extraction module and the perspective transformation module of the neural network model are applied to determine the perspective matrix corresponding to the image features, including: based on the interpolated and scaled image features, applying the fourth feature extraction module to determine the image features converted by the fourth feature extraction module; inputting the image features converted by the fourth feature extraction module into the perspective transformation module, and obtaining the perspective matrix through processing of a two-dimensional convolution layer, a two-dimensional pooling layer, a folding transformation into a one-dimensional layer, and a linear layer.

[0009] In one embodiment of the present invention, after determining the processed pedestrian image, the method further includes: applying a second neural network model based on the pedestrian image to determine pedestrian features and pedestrian category features corresponding to the pedestrian image.

[0010] Correspondingly, the present invention provides a pedestrian image preprocessing device based on pedestrian re-identification, including: a first processing module, used to apply the first feature extraction module of the neural network model based on the pedestrian image to be processed to determine the image features corresponding to the pedestrian image; a second processing module, used to apply the interpolation and scaling algorithm based on the image features to determine the interpolated and scaled image features; a first determination module, used to apply the second feature extraction module and the third feature extraction module of the neural network model based on the interpolated and scaled image features to determine the scaled image; and based on the interpolated and scaled image features, apply the fourth feature extraction module and the perspective transformation module of the neural network model to determine the perspective matrix corresponding to the image features; a second determination module, used to apply the perspective transformation method based on the scaled image and the perspective matrix to determine the processed pedestrian image.

[0011] In one embodiment of the present invention, the input channel value of the processed image of the first feature extraction module is less than the output channel value of the processed image of the first feature extraction module; the input channel value of the processed image of the second feature extraction module is equal to the output channel value of the processed image of the second feature extraction module; the input channel value of the processed image of the third feature extraction module is greater than the output channel value of the processed image of the third feature extraction module; the input channel value of the processed image of the fourth feature extraction module is equal to the output channel value of the processed image of the fourth feature extraction module.

[0012] In one embodiment of the present invention, the first determination module is used to: apply the fourth feature extraction module based on the interpolated and scaled image features to determine the image features converted by the fourth feature extraction module; input the image features converted by the fourth feature extraction module into the perspective transformation module, and obtain the perspective matrix after processing through a two-dimensional convolution layer, a two-dimensional pooling layer, a folding transformation into a one-dimensional layer, and a linear layer.

[0013] In one embodiment of the present invention, the second determination module is further configured to: apply a second neural network model based on the pedestrian image to determine pedestrian features and pedestrian category features corresponding to the pedestrian image.

[0014] The present invention provides a storage medium storing a computer program, which, when executed by a processor, implements the above-mentioned pedestrian image preprocessing method based on pedestrian re-identification.

[0015] The present invention provides a pedestrian image preprocessing system, comprising a memory for storing a computer program; and a processor for running the computer program to implement the above-mentioned pedestrian image preprocessing method based on pedestrian re-identification.

[0016] As described above, the pedestrian image preprocessing method and device based on pedestrian re-identification of the present invention have the following beneficial effects:

[0017] (1) Introduce the interpolation and scaling algorithm into the neural network to reduce the style influence caused by image preprocessing.

[0018] (2) Performing perspective transformation processing in the neural network reduces the differences in pedestrian images under different viewing angles and plays a role in normalizing the semantic information of pedestrians.

[0019] (3) The de-stylization of pedestrian images is achieved, ensuring the uniformity of pedestrian styles and alignment of pedestrian semantics across images, enhancing the robustness and generalization ability of the pedestrian re-identification system, reducing the complexity of the system, and improving the portability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1FIG. 1 is a flowchart of a pedestrian image pre-processing method based on pedestrian re-identification according to an embodiment of the present invention.

[0021] Figure 2 Shown is a neural network model processing diagram of a pedestrian image preprocessing method based on pedestrian re-identification in one embodiment of the present invention.

[0022] Figure 3 FIG. 1 is a schematic structural diagram of a pedestrian image pre-processing device based on pedestrian re-identification according to an embodiment of the present invention.

[0023] Figure 4 Shown is a diagram of a pedestrian image preprocessing system in an embodiment of a pedestrian image preprocessing device based on pedestrian re-identification according to the present invention.

[0024] Component number description

[0025] 31 First processing module

[0026] 32 Second processing module

[0027] 33 First determination module

[0028] 34 Second determination module

[0029] 41 processors

[0030] 42 Memory DETAILED DESCRIPTION

[0031] The following describes the embodiments of the present invention through specific examples. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. The details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other unless they conflict.

[0032] It should be noted that the illustrations provided in the following embodiments are merely schematic illustrations of the basic concept of the present invention. Therefore, the illustrations only show components related to the present invention and are not drawn according to the number, shape, and size of components in actual implementation. In actual implementation, the type, quantity, and proportion of each component may be changed arbitrarily, and the component layout may also be more complex.

[0033] The pedestrian image preprocessing method and device based on pedestrian re-identification of the present invention can introduce the interpolation and scaling algorithm into the neural network to reduce the style influence brought by image preprocessing; at the same time, perspective transformation processing is performed in the neural network to reduce the differences brought by pedestrian images under different viewing angles, and play a role in normalizing the semantic information of pedestrians; in addition, the de-stylization of pedestrian images is also realized to ensure the uniformity of pedestrian styles and alignment of pedestrian semantics across images, enhance the robustness and generalization ability of the pedestrian re-identification system, reduce the complexity of the system, and improve the portability of the system.

[0034] like Figure 1 As shown, in this embodiment, the pedestrian image preprocessing method based on pedestrian re-identification of the present invention includes the following steps:

[0035] Step S1: Based on the pedestrian image to be processed, apply the first feature extraction module of the neural network model to determine the image features corresponding to the pedestrian image.

[0036] Specifically, after acquiring a pedestrian image to be processed through a camera, the first feature extraction module of the neural network model is applied to determine the image features corresponding to the pedestrian image. The image features here are shallow neural network features. For example, the structure of the first feature extraction module is shown in Table 1.

[0037] Table 1

[0038] The first feature extraction module 1.conv2d(kennel_size=7×7,ch_in=3,ch_out=16, stride=1, padding=3),ReLU 2.conv2d(kennel_size=1×1, ch_in=16, ch_out=16, stride=1, padding=0), ReLU 3.BatchNorm2d(ch_in=16,ch_out=16)

[0039] Among them, conv2d represents a two-dimensional convolutional layer, kennel_size represents the convolution kernel size, ch_in is the input channel size, ch_out is the output channel size, stride=1 represents the step size of 2, padding=1 represents the padding size of 1, ReLU represents the activation function, and BatchNorm2d represents two-dimensional batch normalization.

[0040] Step S2: Based on the image features, apply an interpolation and scaling algorithm to determine the features of the image after interpolation and scaling.

[0041] For example, if the image feature dimension is 16*width*height, an interpolation and scaling algorithm is applied to each width*height feature to determine the interpolated and scaled image feature. The interpolation and scaling algorithm can include any of nearest neighbor interpolation, bilinear interpolation, and bicubic interpolation. By introducing the interpolation and scaling algorithm into the neural network to process image features, the neural network can learn and adjust the interpolated and scaled pedestrian features, reducing the impact of uniform input size during neural network preprocessing on the pedestrian image style. Furthermore, the scaled image size can be set according to actual needs.

[0042] Step S3: Based on the interpolated and scaled image features, the second feature extraction module and the third feature extraction module of the neural network model are applied to determine the scaled image; and based on the interpolated and scaled image features, the fourth feature extraction module and the perspective transformation module of the neural network model are applied to determine the perspective matrix corresponding to the image features.

[0043] Specifically, the image features obtained through the interpolation and scaling algorithm contain the variability characteristics of traditional interpolation algorithms. The second and fourth feature extraction modules of the neural network model process these image features, which contain the variability characteristics of traditional interpolation algorithms. The second feature extraction module is primarily used to learn and extract information about these image features. Furthermore, due to the implicit learning capabilities of the neural network model, the second feature extraction module can learn the styles of pedestrian images from different cameras and different scenes during image feature processing, to a certain extent unifying the visual characteristics of the input for subsequent pedestrian re-identification processing and achieving cross-domain image transformation. The fourth feature extraction module is primarily used to transform the information of these image features in preparation for subsequent perspective transformation processing. For example, the structure of the second feature extraction module is shown in Table 2.

[0044] Table 2

[0045] Second feature extraction module 1.conv2d(kennel_size=3×3,ch_in=16,ch_out=16, stride=1, padding=1) 2.BatchNorm2d(kennel_size=3×3, ch_in=16, ch_out=16), ReLU 3.conv2d(kennel_size=3×3,ch_in=3,ch_out=16, stride=1, padding=1) 4. BatchNorm2d (ch_in = 16, ch_out = 16) + input of the second feature extraction module

[0046] Among them, conv2d represents a two-dimensional convolutional layer, kennel_size represents the convolution kernel size, ch_in is the input channel size, ch_out is the output channel size, stride=1 represents the step size of 2, padding=1 represents the padding size of 1, ReLU represents the activation function, and BatchNorm2d represents two-dimensional batch normalization.

[0047] The structure of the fourth feature extraction module is shown in Table III.

[0048] Table 3

[0049] Fourth feature extraction module 1.conv2d(kennel_size=3×3,ch_in=16,ch_out=16, stride=1, padding=1) 2.BatchNorm2d(kennel_size=3×3, ch_in=16, ch_out=16), ReLU 3.conv2d(kennel_size=3×3,ch_in=3,ch_out=16, stride=1, padding=1) 4. BatchNorm2d (ch_in = 16, ch_out = 16) + input of the fourth feature extraction module

[0050] Among them, conv2d represents a two-dimensional convolutional layer, kennel_size represents the convolution kernel size, ch_in is the input channel size, ch_out is the output channel size, stride=1 represents the step size of 2, padding=1 represents the padding size of 1, ReLU represents the activation function, and BatchNorm2d represents two-dimensional batch normalization.

[0051] More specifically, based on the interpolated and scaled image features, the fourth feature extraction module is applied to determine the image features converted by the fourth feature extraction module; the image features converted by the fourth feature extraction module are input into the perspective transformation module, and the perspective matrix is ​​obtained after processing by a two-dimensional convolution layer, a two-dimensional pooling layer, a folding transformation into a one-dimensional layer, and a linear layer.

[0052] More specifically, the structure of the third feature extraction module is shown in Table 4.

[0053] Table 4

[0054] The third feature extraction module 1.conv2d(kennel_size=7×7,ch_in=16,ch_out=3, stride=1, padding=3)

[0055] Among them, conv2d represents a two-dimensional convolutional layer, kennel_size represents the convolution kernel size, ch_in is the input channel size, ch_out is the output channel size, stride=1 represents the step size of 2, and padding=1 represents the padding size of 1.

[0056] More specifically, the input channel value of the image processed by the first feature extraction module is less than the output channel value of the image processed by the first feature extraction module; the input channel value of the image processed by the second feature extraction module is equal to the output channel value of the image processed by the second feature extraction module; the input channel value of the image processed by the third feature extraction module is greater than the output channel value of the image processed by the third feature extraction module; and the input channel value of the image processed by the fourth feature extraction module is equal to the output channel value of the image processed by the fourth feature extraction module. The input channel value in the first feature extraction module is less than the output channel value to achieve dimensionality increase processing of the pedestrian image, and then through processing by the second feature extraction module and the fourth feature extraction module, the learning ability of the neural network for pedestrian images is improved; the input channel value in the third feature extraction module is greater than the output channel value to achieve dimensionality reduction processing of image features.

[0057] More specifically, the structure of the perspective transformation module is shown in Table 5.

[0058] Table 5

[0059] Perspective Transformation Module 1.conv2d(kennel_size=5×5,ch_in=16,ch_out=16, stride=1, padding=1) 2.MaxPool2d(kennel_size=4×4, stride=2), ReLU 3.flatten(ch_in=N,ch_out=N) 4.Linear(ch_in=N,ch_out=9)

[0060] Among them, conv2d represents a two-dimensional convolutional layer, kennel_size represents the convolution kernel size, ch_in is the input channel size, ch_out is the output channel size, stride=1 represents a step size of 2, padding=1 represents a padding size of 1, MaxPool2d represents a two-dimensional pooling layer, ReLU represents an activation function, flatten represents folding the input feature map into a one-dimensional array, and Linear represents a linear layer.

[0061] For example, the perspective matrix obtained after processing by the perspective transformation module is:

[0062]

[0063] Step S4: Based on the scaled image and the perspective matrix, a perspective transformation method is applied to determine a processed pedestrian image. This processing method can obtain a processed pedestrian image that preserves the integrity of pedestrian semantic information while making pedestrian images of different styles and misaligned pedestrian images consistent.

[0064] More specifically, after determining the processed pedestrian image, the method further includes: applying a second neural network model based on the pedestrian image to determine pedestrian features and pedestrian category features corresponding to the pedestrian image.

[0065] For example, after determining the processed pedestrian image, the pedestrian image is input into the second neural network model for pedestrian re-identification for processing, wherein the second neural network model has two loss functions, the first loss function corresponds to the training processing of pedestrian features, and the second loss function corresponds to the training processing of pedestrian category features.

[0066] The processing of steps S1 to S4 above can introduce the interpolation and scaling algorithm into the neural network, reducing the style influence caused by image preprocessing; at the same time, perspective transformation processing is performed in the neural network, which reduces the differences in pedestrian images under different viewing angles and plays a role in normalizing the semantic information of pedestrians; in addition, the de-stylization of pedestrian images is achieved, ensuring the uniformity of pedestrian styles and alignment of pedestrian semantics across images, enhancing the robustness and generalization ability of the pedestrian re-identification system, reducing the complexity of the system, and improving the portability of the system.

[0067] like Figure 3 As shown, in this embodiment, the pedestrian image preprocessing device based on pedestrian re-identification of the present invention includes:

[0068] A first processing module 31 is configured to apply a first feature extraction module of a neural network model to a pedestrian image to be processed to determine image features corresponding to the pedestrian image;

[0069] A second processing module 32 is configured to apply an interpolation and scaling algorithm based on the image features to determine features of the image after interpolation and scaling;

[0070] a first determination module 33 configured to determine a scaled image based on the interpolated and scaled image features by applying the second and third feature extraction modules of the neural network model; and to determine a perspective matrix corresponding to the image features by applying the fourth feature extraction module and the perspective transformation module of the neural network model based on the interpolated and scaled image features;

[0071] The second determining module 34 is configured to determine a processed pedestrian image by applying a perspective transformation method based on the scaled image and the perspective matrix.

[0072] Among them, the input channel value of the processed image of the first feature extraction module is less than the output channel value of the processed image of the first feature extraction module; the input channel value of the processed image of the second feature extraction module is equal to the output channel value of the processed image of the second feature extraction module; the input channel value of the processed image of the third feature extraction module is greater than the output channel value of the processed image of the third feature extraction module; the input channel value of the processed image of the fourth feature extraction module is equal to the output channel value of the processed image of the fourth feature extraction module.

[0073] The first determination module is used to: apply the fourth feature extraction module based on the interpolated and scaled image features to determine the image features converted by the fourth feature extraction module; input the image features converted by the fourth feature extraction module into the perspective transformation module, and obtain the perspective matrix after processing through a two-dimensional convolution layer, a two-dimensional pooling layer, a folding conversion to a one-dimensional layer, and a linear layer.

[0074] The technical features specifically implemented in the pedestrian image preprocessing device based on pedestrian re-identification in this embodiment are basically the same as the principles of each step in the pedestrian image preprocessing method based on pedestrian re-identification in Example 1. The technical content that is common between the method and the device will not be repeated.

[0075] The storage medium of the present invention stores a computer program, which, when executed by a processor, implements the above-mentioned pedestrian image preprocessing method based on pedestrian re-identification.

[0076] like Figure 4 As shown, in this embodiment, the pedestrian image pre-processing system of the present invention includes: a processor 41 and a memory 42 .

[0077] The memory 42 is used to store computer programs.

[0078] The memory 42 includes various media capable of storing program codes, such as ROM, RAM, magnetic disk, USB flash drive, memory card or optical disk.

[0079] The processor 41 is connected to the memory 42 and is configured to execute the computer program stored in the memory 42 so as to enable the pedestrian image preprocessing system to perform the above-mentioned pedestrian image preprocessing method based on pedestrian re-identification.

[0080] Preferably, the above-mentioned processor 41 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0081] In summary, the pedestrian image preprocessing method and device based on pedestrian re-identification of the present invention can introduce an interpolation and scaling algorithm into a neural network, reducing the style influence caused by image preprocessing. At the same time, perspective transformation processing is performed in the neural network, reducing the differences in pedestrian images under different viewing angles, and playing a role in normalizing pedestrian semantic information. In addition, the pedestrian image is de-stylized, ensuring the uniformity of pedestrian style and alignment of pedestrian semantics across image domains, enhancing the robustness and generalization ability of the pedestrian re-identification system, reducing the complexity of the system, and improving the system's portability. Therefore, the present invention effectively overcomes the various shortcomings of the existing technology and has high industrial utilization value.

[0082] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the present invention. Anyone skilled in the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by one of ordinary skill in the art without departing from the spirit and technical principles disclosed herein are intended to be covered by the claims of the present invention.

Claims

1. A pedestrian image preprocessing method based on pedestrian re-identification, characterized in that: The pedestrian image preprocessing method comprises the following steps: Based on a pedestrian image to be processed, applying a first feature extraction module of a neural network model to determine image features corresponding to the pedestrian image; wherein an input channel value of the processed image by the first feature extraction module is less than an output channel value of the processed image by the first feature extraction module; Based on the image features, applying an interpolation and scaling algorithm to determine features of the image after interpolation and scaling; Based on the interpolated and scaled image features, applying a second feature extraction module and a third feature extraction module of a neural network model to determine a scaled image; wherein an input channel value of the image processed by the second feature extraction module is equal to an output channel value of the image processed by the second feature extraction module; and an input channel value of the image processed by the third feature extraction module is greater than an output channel value of the image processed by the third feature extraction module; Based on the interpolated and scaled image features, a fourth feature extraction module and a perspective transformation module of a neural network model are applied to determine a perspective matrix corresponding to the image features; including: based on the interpolated and scaled image features, applying the fourth feature extraction module to determine the image features transformed by the fourth feature extraction module; inputting the image features transformed by the fourth feature extraction module into the perspective transformation module, and obtaining the perspective matrix through processing of a two-dimensional convolution layer, a two-dimensional pooling layer, a folding conversion to a one-dimensional layer, and a linear layer; wherein the input channel value of the processed image of the fourth feature extraction module is equal to the output channel value of the processed image of the fourth feature extraction module; Based on the scaled image and the perspective matrix, a perspective transformation method is applied to determine a processed pedestrian image.

2. The method according to claim 1, characterized in that After determining the processed pedestrian image, the method further includes: Based on the pedestrian image, a second neural network model is applied to determine pedestrian features and pedestrian category features corresponding to the pedestrian image.

3. A pedestrian image preprocessing device based on pedestrian re-identification, characterized in that: include a first processing module configured to apply a first feature extraction module of a neural network model to a pedestrian image to be processed to determine image features corresponding to the pedestrian image; wherein an input channel value of the image processed by the first feature extraction module is smaller than an output channel value of the image processed by the first feature extraction module; A second processing module is configured to apply an interpolation and scaling algorithm based on the image features to determine features of the image after interpolation and scaling; a first determining module configured to determine a scaled image based on the interpolated scaled image features by applying a second feature extraction module and a third feature extraction module of a neural network model; wherein an input channel value of the image processed by the second feature extraction module is equal to an output channel value of the image processed by the second feature extraction module; and an input channel value of the image processed by the third feature extraction module is greater than an output channel value of the image processed by the third feature extraction module; Based on the interpolated and scaled image features, a fourth feature extraction module and a perspective transformation module of a neural network model are applied to determine a perspective matrix corresponding to the image features; including: based on the interpolated and scaled image features, applying the fourth feature extraction module to determine the image features transformed by the fourth feature extraction module; inputting the image features transformed by the fourth feature extraction module into the perspective transformation module, and obtaining the perspective matrix through processing of a two-dimensional convolution layer, a two-dimensional pooling layer, a folding conversion to a one-dimensional layer, and a linear layer; wherein the input channel value of the processed image of the fourth feature extraction module is equal to the output channel value of the processed image of the fourth feature extraction module; The second determining module is configured to determine a processed pedestrian image by applying a perspective transformation method based on the scaled image and the perspective matrix.

4. The device according to claim 3, characterized in that The second determining module is further configured to: Based on the pedestrian image, a second neural network model is applied to determine pedestrian features and pedestrian category features corresponding to the pedestrian image.

5. A storage medium storing program instructions, wherein: When the program instructions are executed, the steps of the pedestrian image preprocessing method based on pedestrian re-identification as described in any one of claims 1 to 2 are implemented.

6. A pedestrian image preprocessing system, characterized by: The method comprises a memory for storing a computer program and a processor for running the computer program to implement the steps of the pedestrian image preprocessing method based on pedestrian re-identification according to any one of claims 1 to 2.

Images