A pedestrian re-identification method and system based on fisheye image correction based on diffusion model

Abstract

The present invention discloses a method and system for pedestrian re-identification based on fisheye image correction using a diffusion model. The method comprises: a correction step, constructing a fisheye pedestrian image correction network based on a diffusion model for fisheye pedestrian image dedistortion; a model training step, constructing a dual-branch deep learning network, jointly using the corrected pedestrian image and the fisheye pedestrian image to train a pedestrian re-identification model, fully utilizing the pedestrian image information to improve re-identification accuracy; and a re-identification image acquisition step, using the acquired pedestrian re-identification model to extract features from the query image and the registered image, respectively selecting the registered image with a close feature distance to the query image. The present invention can solve the problem of pedestrian image distortion and matching difficulty in fisheye lens images, thereby improving the accuracy of pedestrian re-identification in fisheye images and the visual comfort of observation and analysis.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and in particular to a pedestrian re-identification method and system based on diffusion model-corrected fisheye images. Background Art

[0002] Pedestrian re-identification (PDR) utilizes multiple conventional cameras with non-overlapping perspectives to search and identify different pedestrians, thereby matching and identifying them. It is widely used in intelligent security, criminal investigation, and other areas. Conventional cameras have a narrow viewing angle, necessitating the installation of multiple cameras for wide-angle, all-around monitoring. Using fisheye lens monitoring can cover the perspectives of multiple conventional cameras, facilitating PDR in large public spaces. However, the strong distortion and image aberration caused by fisheye lenses hinders the accuracy of PDR and the visual observation required for analysis. Therefore, correcting fisheye pedestrian images to improve the accuracy and visual comfort of PDR is a key issue in PDR. Summary of the Invention

[0003] In response to the problems of the existing technology, the purpose of the present invention is to provide a pedestrian re-identification method and system based on diffusion model correction of fisheye images, which can solve the problems of pedestrian image distortion and difficulty in matching through fisheye lenses, and improve the accuracy of pedestrian re-identification in fisheye images and the visual comfort of analysis and observation.

[0004] The present invention adopts the following technical solutions:

[0005] On the one hand, a pedestrian re-identification method based on fisheye image correction based on a diffusion model includes:

[0006] A correction step, constructing a fisheye pedestrian image correction network based on a diffusion model based on the conventional pedestrian image and the fisheye pedestrian image, so as to correct the input fisheye pedestrian image and obtain a corrected pedestrian image;

[0007] In the model training step, a two-branch deep learning network is constructed, and the rectified pedestrian images and fisheye pedestrian images are jointly used to train the pedestrian re-identification model to obtain a trained pedestrian re-identification model;

[0008] The re-identification image acquisition step is to input the registered fisheye pedestrian image into the fisheye pedestrian image correction network to obtain the registered corrected pedestrian image; use the trained pedestrian re-identification model to extract features of the query pedestrian image, the registered fisheye pedestrian image and the registered corrected pedestrian image, and select the preset registered fisheye pedestrian images with the closest feature distance to the query pedestrian image and the preset registered corrected pedestrian images with the closest feature distance to the query pedestrian image as the pedestrian re-identification results.

[0009] Preferably, the diffusion model is a pre-trained latent diffusion model LDM.

[0010] Preferably, the correction step specifically includes:

[0011] Diffusion process: The encoder encodes the regular pedestrian image v to obtain the potential vector z0 = Encoder(v), and then adds Gaussian noise ∈1,…,∈ t ,…,∈ T , obtain the potential vector z1,…z t ,…,z T ; Gaussian noise∈ t The normalized latent vector z t-1 It is obtained through random sampling and is specifically expressed as follows:

[0012] ∈ t =Sample(Norm(z t-1 )),t=1,2,3...,T (1)

[0013] Among them, Sample represents random sampling operation, and Norm represents normalization operation;

[0014] The process of adding Gaussian noise is expressed as the following formula (2):

[0015]

[0016] Among them, α t represents the noise weight;

[0017] Inverse diffusion process: construct denoising module ∈ θ , the latent vector z t As input, the fisheye pedestrian image is used as conditional information to predict z t The noise added relative to z0, so as to obtain z t-1 The conditional probability distribution of is expressed as follows:

[0018]

[0019] in, Indicates that noise is added to the potential vector z0 to generate z t The noise weight of

[0020] After iterative training until the loss function reaches convergence, the denoising module ∈ θ The prediction noise of is close to Gaussian noise ∈, thus simulating the inverse process of adding noise in the diffusion process and obtaining the fisheye pedestrian image correction network;

[0021] Correction process: Use the fisheye pedestrian image correction network trained in the back-diffusion process, take the fisheye pedestrian image as input, and obtain the corrected pedestrian image.

[0022] Preferably, the model training step specifically includes:

[0023] Construct a two-branch deep learning network Net, where the first branch Net1 is used to learn the features of fisheye pedestrian images, and the second branch Net2 is used to learn the features of corrected pedestrian images;

[0024] Select the fisheye pedestrian image training set X and the corresponding corrected pedestrian image training set Y obtained by the fisheye pedestrian image correction network, and iteratively train the dual-branch deep learning network Net until the loss function converges to obtain a trained pedestrian re-identification model. Its loss function is expressed as formula (4):

[0025]

[0026] Where X represents the fisheye pedestrian image training set; Y represents the corrected pedestrian image training set; represents the pedestrian re-identification loss function from the fisheye pedestrian image to the corrected pedestrian image; represents the pedestrian re-identification loss function for correcting pedestrian images to fisheye pedestrian images; represents the pedestrian re-identification loss function from fisheye pedestrian image to fisheye pedestrian image; Represents the pedestrian re-identification loss function from the rectified pedestrian image to the rectified pedestrian image.

[0027] Preferably, the person re-identification loss function is expressed as follows:

[0028]

[0029]

[0030]

[0031]

[0032] Among them, dist represents the calculated distance; x a Anchor point samples representing fisheye pedestrian image features; x n Represents the positive sample of fisheye pedestrian image features; x p Represents the negative sample of fisheye pedestrian image features; y a Represents the anchor point sample for correcting pedestrian image features; y n represents the positive sample of the corrected pedestrian image features; y p represents the negative sample for correcting pedestrian image features; ρ>0 represents the manually set interval constant.

[0033] Preferably, the step of obtaining the re-identified image specifically includes:

[0034] Use the fisheye pedestrian image correction network to obtain the registered and corrected pedestrian image test set corresponding to the registered fisheye pedestrian image;

[0035] Use the first branch Net1 to extract the query fisheye pedestrian image feature f q X and registered fisheye pedestrian image features Use the second branch Net2 to extract the registered and corrected pedestrian image features K represents the number of registered fisheye pedestrian image sets;

[0036] Calculate the query fisheye pedestrian image features f in sequence q X And each registered corrected pedestrian image feature distance Calculate and query the fisheye pedestrian image feature f q X And each registered fisheye pedestrian image feature distance

[0037] The feature distance D X→Y Sort from small to large to get the distance sorted list List X→Y , select the registered and corrected pedestrian images corresponding to the first M distances in the list; set the feature distance D X→X Sort from small to large to get the distance sorted list List X→X , select the registered fisheye pedestrian images corresponding to the first M distances in the list; output the re-identification results of the selected registered corrected pedestrian images and the registered fisheye pedestrian images; where M≤K.

[0038] On the other hand, a pedestrian re-identification system based on fisheye image correction based on a diffusion model includes:

[0039] A correction module is used to construct a fisheye pedestrian image correction network based on a diffusion model based on a conventional pedestrian image and a fisheye pedestrian image, so as to correct the input fisheye pedestrian image and obtain a corrected pedestrian image;

[0040] The model training module is used to construct a two-branch deep learning network and jointly use the rectified pedestrian images and fisheye pedestrian images to train the pedestrian re-identification model to obtain a trained pedestrian re-identification model;

[0041] The re-identification image selection module is used to input the registered fisheye pedestrian image into the fisheye pedestrian image correction network to obtain the registered corrected pedestrian image; use the trained pedestrian re-identification model to extract features of the query pedestrian image, the registered fisheye pedestrian image and the registered corrected pedestrian image, and select the preset registered fisheye pedestrian images with the closest feature distance to the query pedestrian image and the preset registered corrected pedestrian images with the closest feature distance to the query pedestrian image as the pedestrian re-identification results.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] 1. The present invention can solve the problem of pedestrian image distortion and difficulty in matching under fisheye lens, and improve the accuracy of pedestrian re-identification in fisheye images and the visual comfort of observation. Specifically, the fisheye pedestrian image is first dedistorted through a fisheye pedestrian image correction network; then a dual-branch deep learning network is constructed to jointly use the corrected pedestrian image and the fisheye pedestrian image to train a pedestrian re-identification model to fully utilize the pedestrian image information to improve the re-identification accuracy; finally, the pedestrian re-identification model is used to extract features from the query image and the registered image, and the registered image with the closest feature distance to the query image is selected;

[0044] 2. The present invention can use the corrected pedestrian image corresponding to each fisheye pedestrian image as the re-identification result, which is convenient for users to observe.

[0045] 3. The present invention can be applied to intelligent video surveillance systems in smart cities, safe cities, and smart transportation, and has wide application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a flow chart of a method for pedestrian re-identification based on fisheye image correction based on a diffusion model according to an embodiment of the present invention;

[0047] Figure 2 Schematic diagram of a fisheye image correction network according to an embodiment of the present invention;

[0048] Figure 3 Schematic diagram of model training steps in an embodiment of the present invention;

[0049] Figure 4 A schematic diagram of the steps for obtaining a re-identified image according to an embodiment of the present invention;

[0050] Figure 5 This is a structural block diagram of a pedestrian re-identification system based on fisheye image correction based on a diffusion model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0051] Below in conjunction with specific embodiment, further set forth the present invention.Should be understood that these embodiments are only used to illustrate the present invention and are not used in limiting the scope of the present invention.In addition, should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall equally within the scope limited by the appended claims of the application.

[0052] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present invention and the features therein can be combined with each other without conflict.

[0053] The following description sets forth numerous specific details to facilitate a thorough understanding of the present invention. The embodiments described are merely some, not all, of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are intended to fall within the scope of protection of the present invention.

[0054] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0055] The terms "first," "second," and "third," etc., in the specification and claims of the present invention and the accompanying drawings are used to distinguish between different items, not to describe a specific order. Furthermore, the terms "comprise," "comprising," and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or elements is not limited to the listed steps or elements but may optionally include steps or elements not listed, or may optionally include other steps or elements inherent to the process, method, product, or apparatus.

[0056] See also Figure 1 As shown, this embodiment provides a pedestrian re-identification method based on diffusion model correction of fisheye images, including the following steps:

[0057] Correction step 1: constructing a fisheye pedestrian image correction network based on a diffusion model based on the conventional pedestrian image and the fisheye pedestrian image to correct the input fisheye pedestrian image to obtain a corrected pedestrian image;

[0058] In the second model training step, a two-branch deep learning network is constructed to jointly use the rectified pedestrian images and the fisheye pedestrian images to train the pedestrian re-identification model to obtain a trained pedestrian re-identification model.

[0059] In the re-identification image acquisition step 3, the registered fisheye pedestrian image is input into the fisheye pedestrian image correction network to obtain the registered corrected pedestrian image; the trained pedestrian re-identification model is used to extract features of the query pedestrian image, the registered fisheye pedestrian image and the registered corrected pedestrian image, and a preset number of registered fisheye pedestrian images with the closest feature distance to the query pedestrian image and a preset number of registered corrected pedestrian images with the closest feature distance to the query pedestrian image are selected as the pedestrian re-identification results.

[0060] The present invention corrects the fisheye image based on the diffusion model, and can simultaneously return the fisheye image and the corrected image as the fisheye image pedestrian re-identification result, which is convenient for human visual observation.

[0061] In this embodiment, the diffusion model in step 1 correction step is a pre-trained latent diffusion model (Latent Diffusion Models, LDM). In this embodiment, see Figure 2 As shown in Figure 1, the pre-trained LDM is fine-tuned using pre-collected fisheye pedestrian images and corresponding regular pedestrian images (taken by a regular camera) to construct a fisheye image correction network. The fine-tuning process includes diffusion, de-diffusion, and correction, as follows:

[0062] Step (1.1) Diffusion process: The encoder encodes the regular pedestrian image v to obtain the potential vector z0 = Encoder(v), and then adds Gaussian noise ∈1,…,∈ t ,…,∈ T Get the latent vector z1,…z t ,…,z T . Gaussian noise ∈ t The normalized latent vector z t-1 It is obtained through random sampling, specifically expressed as:

[0063] ∈ t =Sample(Norm(z t-1 )). (1)

[0064] The process of adding Gaussian noise is expressed as:

[0065]

[0066] Among them, α t Represents the noise weight. Then the noise is added from the latent vector z0 to the latent vector z t The process is expressed as:

[0067]

[0068] in, Indicates that noise is added to the potential vector z0 to generate z tThe noise weight is

[0069]

[0070] Step (1.2) reverse diffusion process: construct denoising module ∈ θ , the latent vector z t As input, the fisheye image is used as conditional information to predict z t The noise added relative to z0, so as to obtain z t-1 The conditional probability distribution of is expressed as:

[0071]

[0072] Thus, the potential vector z is restored by random sampling from the conditional probability distribution t-1 After iterative training until the loss function reaches convergence, the denoising module ∈ θ The predicted noise of is close to Gaussian noise ∈, thus simulating the inverse process of adding noise in the diffusion process of step (1.1) to obtain the fisheye image correction network. In this embodiment, its loss function is expressed as:

[0073]

[0074] Among them, x represents the fisheye pedestrian image, which is used as conditional information here; Encoder θ Represents the conditional information encoder.

[0075] Step (1.3) Correction process: Use the fisheye pedestrian image correction network obtained by training in the back diffusion process, take the fisheye pedestrian image as input, and obtain the corrected pedestrian image for the training step (2).

[0076] In this embodiment, the model training step specifically includes:

[0077] Step (2.1): To avoid the loss of some image information in the correction step, in this embodiment, a deep network (such as VGGNet, ResNet, DenseNet, etc.) is used to construct a two-branch deep learning network Net, see Figure 3 As shown in the figure, the first branch, Net1, is used to learn fisheye pedestrian image features, and the second branch, Net2, is used to learn corrected pedestrian image features. In this embodiment, Net1 and Net2 have independent shallow network parameters and share deep network parameters, avoiding information loss in corrected pedestrian images and learning consistent high-level semantic features.

[0078] Step (2.2): Using the fisheye pedestrian image training set X and the corrected pedestrian image training set Y obtained in step (1.3), iteratively train the two-branch deep learning network Net constructed in step (2.1) until the loss function converges, and obtain the fisheye image corrected pedestrian re-identification model based on the diffusion model. Among them, the fisheye image pedestrian re-identification loss function is expressed as,

[0079]

[0080] Where X represents the fisheye pedestrian image domain; Y represents the corrected pedestrian image domain; and They represent the pedestrian re-identification loss function from the fisheye pedestrian image to the corrected pedestrian image, the pedestrian re-identification loss function from the corrected pedestrian image to the fisheye pedestrian image, the pedestrian re-identification loss function from the fisheye pedestrian image to the fisheye pedestrian image, and the pedestrian re-identification loss function from the corrected pedestrian image to the corrected pedestrian image. In this embodiment, the pedestrian re-identification loss function is calculated as follows:

[0081]

[0082]

[0083]

[0084]

[0085] Wherein, dist represents the calculated distance; x and y represent the fisheye pedestrian image features and the corrected pedestrian image features, respectively; a, p, and n represent the anchor point sample, positive sample, and negative sample, respectively; ρ>0 represents a manually set interval constant, which is set to 1 in this embodiment.

[0086] In this embodiment, see Figure 3 As shown, the steps of re-identifying the image acquisition specifically include:

[0087] Step (3.1): Use the fisheye pedestrian image correction network to obtain the registered corrected pedestrian image test set corresponding to the registered fisheye pedestrian image;

[0088] Step (3.2): Use Net1 to extract the query fisheye pedestrian image feature f q X and registered fisheye pedestrian image features Use Net2 to extract and register pedestrian image features K represents the number of registered fisheye pedestrian image sets.

[0089] Step (3.3): Calculate the query fisheye pedestrian image features f in sequence q XAnd each registered corrected pedestrian image feature distance And each registered fisheye pedestrian image feature distance

[0090] Step (3.4): Sort the feature distances from small to large and obtain the distance sorting list List X→Y and List X→X , respectively select the registered corrected pedestrian images and registered fisheye pedestrian images corresponding to the first M (M≤K) distances in the list, which are the pedestrian re-identification results of the fisheye image corrected based on the diffusion model.

[0091] See also Figure 5 As shown, this embodiment also discloses a pedestrian re-identification system based on fisheye image correction based on a diffusion model, including:

[0092] The correction module 501 is used to construct a fisheye pedestrian image correction network based on a diffusion model based on the conventional pedestrian image and the fisheye pedestrian image, so as to correct the input fisheye pedestrian image and obtain a corrected pedestrian image;

[0093] The model training module 502 is used to construct a dual-branch deep learning network, and jointly use the corrected pedestrian image and the fisheye pedestrian image to train a pedestrian re-identification model to obtain a trained pedestrian re-identification model;

[0094] The re-identification image selection module 503 is used to input the registered fisheye pedestrian image into the fisheye pedestrian image correction network to obtain the registered corrected pedestrian image; use the trained pedestrian re-identification model to extract features of the query pedestrian image, the registered fisheye pedestrian image and the registered corrected pedestrian image, and select the preset registered fisheye pedestrian images closest to the query pedestrian image feature and the preset registered corrected pedestrian images closest to the query pedestrian image feature as the pedestrian re-identification results.

[0095] Specific Implementation of a Person Re-identification System for Correcting Fisheye Images Based on a Diffusion Model The same pedestrian re-identification method based on correcting fisheye images based on a diffusion model will not be repeated in this embodiment.

[0096] It is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-restrictive, and the scope of the invention is defined by the appended claims rather than the above description, and it is intended that all changes that fall within the meaning and scope of the equivalent elements of the claims be included in the present invention. Any figure mark in the claims should not be regarded as limiting the claim involved. In addition, it is obvious that the word "comprising" does not exclude other units or, and the singular does not exclude the plural. Multiple units or systems stated in the system claim may also be implemented by one unit or system through software or hardware.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A pedestrian re-identification method based on fisheye image correction based on diffusion model, characterized in that: include: A correction step, constructing a fisheye pedestrian image correction network based on a diffusion model based on the conventional pedestrian image and the fisheye pedestrian image, so as to correct the input fisheye pedestrian image and obtain a corrected pedestrian image; In the model training step, a two-branch deep learning network is constructed, and the rectified pedestrian images and fisheye pedestrian images are jointly used to train the pedestrian re-identification model to obtain a trained pedestrian re-identification model; In the re-identification image acquisition step, the registered fisheye pedestrian image is input into the fisheye pedestrian image correction network to obtain a registered and corrected pedestrian image; using the trained pedestrian re-identification model, feature extraction is performed on the query pedestrian image, the registered fisheye pedestrian image, and the registered and corrected pedestrian image, and a preset number of registered fisheye pedestrian images with the closest feature distance to the query pedestrian image and a preset number of registered and corrected pedestrian images with the closest feature distance to the query pedestrian image are selected as the pedestrian re-identification results; The correction step specifically includes: Diffusion process: The encoder encodes the regular pedestrian image v to obtain the potential vector z0 = Encoder(v), and then adds Gaussian noise ∈1,…,∈ t ,…,∈ T , obtain the potential vector z1,…z t ,…,z T ; Gaussian noise∈ t The normalized latent vector z t-1 It is obtained through random sampling and is specifically expressed as follows: ∈ t =Sample(Norm(z t-1 )), t=1,2,3...,T (1) Among them, Sample represents random sampling operation, and Norm represents normalization operation; The process of adding Gaussian noise is expressed as the following formula (2): Among them, α t represents the noise weight; Inverse diffusion process: construct denoising module ∈ θ , the latent vector z t As input, the fisheye pedestrian image is used as conditional information to predict z t The noise added relative to z0, so as to obtain z t-1 The conditional probability distribution of is expressed as follows: t=1,2,3,...,T in, Indicates that noise is added to the potential vector z0 to generate z t The noise weight of After iterative training until the loss function reaches convergence, the denoising module ∈ θ The prediction noise of is close to Gaussian noise ∈, thus simulating the inverse process of adding noise in the diffusion process and obtaining the fisheye pedestrian image correction network; Correction process: Use the fisheye pedestrian image correction network trained in the back-diffusion process, take the fisheye pedestrian image as input, and obtain the corrected pedestrian image; The model training steps specifically include: Construct a two-branch deep learning network Net, where the first branch Net1 is used to learn the features of fisheye pedestrian images, and the second branch Net2 is used to learn the features of corrected pedestrian images; Select the fisheye pedestrian image training set X and the corresponding corrected pedestrian image training set Y obtained by the fisheye pedestrian image correction network, and iteratively train the dual-branch deep learning network Net until the loss function converges to obtain a trained pedestrian re-identification model. Its loss function is expressed as formula (4): Where X represents the fisheye pedestrian image training set; Y represents the corrected pedestrian image training set; represents the pedestrian re-identification loss function from the fisheye pedestrian image to the corrected pedestrian image; represents the pedestrian re-identification loss function for correcting pedestrian images to fisheye pedestrian images; represents the pedestrian re-identification loss function from fisheye pedestrian image to fisheye pedestrian image; Represents the pedestrian re-identification loss function from the rectified pedestrian image to the rectified pedestrian image.

2. The pedestrian re-identification method based on diffusion model correction of fisheye images according to claim 1, characterized in that: The diffusion model is a pre-trained latent diffusion model LDM.

3. The pedestrian re-identification method based on diffusion model correction of fisheye images according to claim 1, characterized in that: The pedestrian re-identification loss function is expressed as follows: Among them, dist represents the calculated distance; x a Anchor point samples representing fisheye pedestrian image features; x n Represents the positive sample of fisheye pedestrian image features; x p Represents the negative sample of fisheye pedestrian image features; y a Represents the anchor point sample for correcting pedestrian image features; y n represents the positive sample of the corrected pedestrian image features; y p represents the negative sample for correcting pedestrian image features; ρ>0 represents a manually set interval constant.

4. The pedestrian re-identification method based on diffusion model correction of fisheye images according to claim 1, characterized in that: The step of obtaining the re-identified image specifically includes: Use the fisheye pedestrian image correction network to obtain the registered and corrected pedestrian image test set corresponding to the registered fisheye pedestrian image; Use the first branch Net1 to extract the query fisheye pedestrian image features and registered fisheye pedestrian image features Use the second branch Net2 to extract the registered and corrected pedestrian image features K represents the number of registered fisheye pedestrian image sets; Calculate and query the fisheye pedestrian image features in sequence And each registered corrected pedestrian image feature distance Calculate and query fisheye pedestrian image features And each registered fisheye pedestrian image feature distance The feature distance D X→Y Sort from small to large to get the distance sorted list List X→Y , select the registered and corrected pedestrian images corresponding to the first M distances in the list; set the feature distance D X→X Sort from small to large to get the distance sorted list List X →X , select the registered fisheye pedestrian images corresponding to the first M distances in the list; output the re-identification results of the selected registered corrected pedestrian images and the registered fisheye pedestrian images; where M≤K.

5. A pedestrian re-identification system based on fisheye image correction based on diffusion model, characterized in that: Based on the method according to any one of claims 1 to 4, comprising: A correction module is used to construct a fisheye pedestrian image correction network based on a diffusion model based on a conventional pedestrian image and a fisheye pedestrian image, so as to correct the input fisheye pedestrian image and obtain a corrected pedestrian image; The model training module is used to construct a two-branch deep learning network and jointly use the rectified pedestrian images and fisheye pedestrian images to train the pedestrian re-identification model to obtain a trained pedestrian re-identification model; The re-identification image selection module is used to input the registered fisheye pedestrian image into the fisheye pedestrian image correction network to obtain the registered corrected pedestrian image; use the trained pedestrian re-identification model to extract features of the query pedestrian image, the registered fisheye pedestrian image and the registered corrected pedestrian image, and select the preset registered fisheye pedestrian images with the closest feature distance to the query pedestrian image and the preset registered corrected pedestrian images with the closest feature distance to the query pedestrian image as the pedestrian re-identification results.

Images