A pedestrian re-identification method and system

Abstract

The present application relates to the field of intelligent decision-making, and discloses a pedestrian re-identification method, comprising: preprocessing pedestrian image data to obtain preprocessed images; extracting local features of the preprocessed images to obtain pedestrian local features; extracting global features of the preprocessed images; performing adaptive noise matching processing on the pedestrian local features and the pedestrian global features to convert them into adaptive local vectors and adaptive global vectors; detecting human key points, constructing connection edges between any human key points, constructing a posture node matrix, and constructing a posture edge matrix according to the connection edges; performing update and aggregation processing on the node feature matrix and the edge feature matrix, and performing dimension reduction and splicing processing on the aggregated node matrix and the aggregated edge matrix; fusing the adaptive local vectors and the adaptive global vectors and the aggregated posture vectors, calculating the cosine similarity between a pedestrian fusion vector and a pedestrian feature vector, and determining the identification result of a to-be-identified pedestrian. The present application can improve the pedestrian re-identification accuracy.

Description

Technical Field

[0001] This invention relates to the field of intelligent decision-making, and in particular to a pedestrian re-identification method and system. Background Technology

[0002] Pedestrian re-identification technology is a further extension of facial recognition technology. When faces are blurred or lack facial feature information for auxiliary identification, pedestrian re-identification technology needs to identify specific pedestrians based on a small number of low-quality or occluded pedestrian images. Because pedestrian re-identification emphasizes pedestrian identification, tracking, and prediction across video surveillance, it is mainly applied in outdoor monitored locations such as big data criminal investigation, intelligent security, and intelligent missing person search, and has significant practical implications for public place security management.

[0003] The mainstream datasets for pedestrian re-identification currently include CUHK03, Market1501, DukeMTMC-reID, and MSMT17. These datasets contain a small number of pedestrian images of poor quality, which are usually affected by varying degrees of lighting changes, obstacle occlusion, viewpoint blur, pedestrian pose changes, and background interference. However, existing model algorithms perform poorly on different datasets, resulting in low accuracy in pedestrian re-identification. Summary of the Invention

[0004] This invention provides a pedestrian re-identification method and system, the main purpose of which is to improve the accuracy of pedestrian re-identification.

[0005] To achieve the above objectives, the present invention provides a pedestrian re-identification method, comprising:

[0006] Acquire image data of the pedestrian to be identified, and preprocess the pedestrian image data to obtain a preprocessed image;

[0007] Local feature extraction is performed on the preprocessed image to obtain the local pedestrian features of the pedestrian to be identified;

[0008] Global feature extraction is performed on the preprocessed image to obtain the global pedestrian features of the pedestrian to be identified;

[0009] Adaptive noise matching processing is performed on the pedestrian local features and the pedestrian global features respectively to obtain adaptive local features and adaptive global features, and the adaptive local features and the adaptive global features are transformed into adaptive local vectors and adaptive global vectors;

[0010] Detect human key points in the preprocessed image and construct connecting edges between any of the human key points. Based on the human key points, construct the pose node matrix of the pedestrian to be identified and the pose edge matrix of the pedestrian to be identified based on the connecting edges.

[0011] The node feature matrix and the edge feature matrix are updated and aggregated to obtain an aggregated node matrix and an aggregated edge matrix. The aggregated node matrix and the aggregated edge matrix are then reduced in dimension and concatenated to obtain an aggregated pose vector.

[0012] By fusing the adaptive local vector, the adaptive global vector, and the aggregated pose vector, a pedestrian fusion vector is obtained. The cosine similarity between the pedestrian fusion vector and the pedestrian feature vector in the pre-built pedestrian database is calculated. Based on the cosine similarity, the recognition result of the pedestrian to be identified is determined.

[0013] Optionally, the step of extracting local features from the preprocessed image to obtain the local pedestrian features of the pedestrian to be identified includes:

[0014] Identify the structural locations of pedestrian bodies in the preprocessed image;

[0015] Based on the structural location, the preprocessed image is segmented to obtain segmented regions;

[0016] Extract the regional features of the segmented region and stitch the regional features together to obtain the stitched local features of the preprocessed image. Use the stitched local features as the pedestrian local features.

[0017] Optionally, the step of segmenting the preprocessed image into segmented regions based on the structural location includes:

[0018] Construct the bounding rectangle of the structure location;

[0019] The segmentation region of the preprocessed image is determined based on the circumscribed rectangle.

[0020] Optionally, the adaptive noise matching processing performed on the pedestrian local features and the pedestrian global features to obtain adaptive local features and adaptive global features includes:

[0021] Obtain the segmented regions corresponding to the local features of the pedestrian and the preprocessed images corresponding to the global features of the pedestrian;

[0022] Calculate the adaptive noise ratio of the segmented region to obtain the segmented region noise ratio, and calculate the adaptive noise ratio of the preprocessed image to obtain the global image noise ratio;

[0023] Based on the noise ratio of the segmented region and the noise ratio of the global image, adaptive noise processing is performed on the segmented region and the preprocessed image respectively to obtain a noisy segmented image and a noisy global image.

[0024] Feature extraction is performed on the noise segmentation image and the noise global image to obtain adaptive local features and adaptive global features.

[0025] Optionally, calculating the adaptive noise ratio of the segmented region and the preprocessed image includes:

[0026] The adaptive noise ratio of the segmented region is calculated using the following formula to obtain the noise ratio of the segmented region:

[0027]

[0028] Where β represents the segmentation region noise ratio, i represents the segmentation region of the pedestrian, j is a segmentation region different from i, k represents the number of matching layer weights in the adaptive noise matching network, l is the number of hidden layers, f(·) is the activation function, and x i y represents the input of the i-th segmentation region in the adaptive noise matching network. i Let W represent the output of the i-th segmentation region in the adaptive noise matching network, W be the weight parameter, λ be the regularization penalty coefficient, α be the adaptive noise adjustment factor, and θ be the value of the segmentation. i Let N be the adaptive noise matrix, and let N represent the total number of segmented regions for the same pedestrian.

[0029] Optionally, detecting human key points in the preprocessed image includes:

[0030] The pre-trained human keypoint detection model is used to detect keypoints in the preprocessed image to obtain the detected keypoints.

[0031] The key detection points are marked to obtain the marking sequence number;

[0032] The human body key points are generated based on the detected key points and the marked sequence number.

[0033] Optionally, constructing the pose node matrix of the pedestrian to be identified based on the human body key points includes:

[0034] Extract the node features of the key points of the human body;

[0035] Based on the node features and the labeling sequence number of the human body key points, construct the node feature vector of the human body key points;

[0036] The node feature vectors are vertically concatenated according to the ascending order of the marker numbers of the human body key points to obtain the posture node matrix.

[0037] Optionally, constructing the pose edge matrix of the pedestrian to be identified based on the connecting edges includes:

[0038] Identify the marker numbers of the key human body points corresponding to the connecting edges;

[0039] Calculate the length of the connecting edge using the following formula;

[0040]

[0041] Where, d ij x represents the length of the edge connecting the two key points of the human body, corresponding to the i and j-th points. i x j Let x and y represent the x-coordinates of the i-th and j-th key points of the human body, respectively. i y j Let i and j represent the ordinates of the i-th and j-th human body key points, respectively, where i and j represent the marker numbers of the human body key points, and k represents the number of human body key points.

[0042] The attitude edge matrix is ​​constructed based on the marker number and the edge length.

[0043] Optionally, the step of updating and aggregating the attitude node matrix and the attitude edge matrix to obtain an aggregated node matrix and an aggregated edge matrix includes:

[0044] The updated node matrix of the attitude node matrix is ​​calculated using the following formula;

[0045]

[0046]

[0047]

[0048] Among them, M (l+1) Let σ represent the aggregation node matrix, I represent the activation function, W represent the learning weight parameters, l represent the number of convolutional layers, and P~ represent the degree matrix of the edge feature matrix S. The main diagonal elements of the P~ matrix represent the number of edges in each node.

[0049] Calculate the attitude edge matrix of the updated node matrix, and use the attitude edge matrix as the updated edge matrix;

[0050] When the updated node matrix meets the preset conditions, the updated node matrix is ​​used as the aggregated node matrix, and the updated edge matrix corresponding to the updated node matrix is ​​used as the aggregated edge matrix.

[0051] To address the above problems, the present invention also provides a pedestrian re-identification system, the system comprising:

[0052] The image preprocessing module is used to acquire pedestrian image data of the pedestrian to be identified, and to preprocess the pedestrian image data to obtain a preprocessed image;

[0053] The local feature generation module is used to extract local features from the preprocessed image to obtain the local pedestrian features of the pedestrian to be identified.

[0054] A global feature generation module is used to extract global features from the preprocessed image to obtain the global pedestrian features of the pedestrian to be identified.

[0055] An adaptive noise matching module is used to perform adaptive noise matching processing on the pedestrian local features and the pedestrian global features respectively to obtain adaptive local features and adaptive global features, and to convert the adaptive local features and the adaptive global features into adaptive local vectors and adaptive global vectors.

[0056] The pose matrix generation module is used to detect human key points in the preprocessed image, construct connection edges between any human key points, construct the pose node matrix of the pedestrian to be identified based on the human key points, and construct the pose edge matrix of the pedestrian to be identified based on the connection edges.

[0057] The aggregated pose vector generation module is used to update and aggregate the node feature matrix and the edge feature matrix to obtain an aggregated node matrix and an aggregated edge matrix, and to perform dimensionality reduction and concatenation on the aggregated node matrix and the aggregated edge matrix to obtain an aggregated pose vector.

[0058] The recognition result determination module is used to fuse the adaptive local vector, the adaptive global vector, and the aggregated pose vector to obtain a pedestrian fusion vector, calculate the cosine similarity between the pedestrian fusion vector and the pedestrian feature vector in the pre-built pedestrian database, and determine the recognition result of the pedestrian to be identified based on the cosine similarity.

[0059] As can be seen, this embodiment of the invention acquires pedestrian image data of the pedestrian to be identified, preprocesses the pedestrian image data to obtain a preprocessed image that removes background noise, resulting in a background-free pedestrian image. This image serves as the target for subsequent operations. Local feature extraction on the preprocessed image yields local features of the pedestrian after background noise removal, while global feature extraction on the preprocessed image yields global features representing the overall information of the pedestrian. Furthermore, this embodiment of the invention, by performing adaptive noise matching processing on the pedestrian's local and global features respectively, obtains more differentiated and discriminative local and global features, thereby improving the robustness of re-identification. Converting the adaptive local and global features into adaptive local and global vectors reduces the dimensionality of the adaptive local and global features, improving computational efficiency. Detecting key human body points in the preprocessed image allows for the extraction of important features of the human body structure. By constructing connection edges between any of the human body's key points, a relationship can be established between these key points. Based on the human body's key points and the connection edges, the pose node matrix and pose edge matrix of the pedestrian to be identified can be constructed to mine the posture patterns such as body shape and movement of the pedestrian in the image, which can be used as the pose features of the pedestrian to be identified. Furthermore, in this embodiment of the invention, the pose node matrix and the pose edge matrix can be updated and aggregated to enhance the pedestrian's pose features, resulting in an enhanced pose feature expression. By fusing the adaptive local vector, the adaptive global vector, and the aggregated pose vector, a more comprehensive and multi-dimensional pedestrian feature of the pedestrian to be identified can be obtained. Calculating the cosine similarity between the pedestrian fusion vector and the pedestrian feature vector in the pre-constructed pedestrian database can determine the degree of similarity between the pedestrian to be identified and the pedestrians in the pre-constructed pedestrian database. Based on the cosine similarity, the identification result of the pedestrian to be identified can be determined, thus achieving the final pedestrian re-identification. Therefore, the pedestrian re-identification method and system proposed in this embodiment of the invention can improve the accuracy of pedestrian re-identification. Attached Figure Description

[0060] Figure 1 This is a flowchart illustrating a pedestrian re-identification method according to an embodiment of the present invention.

[0061] Figure 2 This is a schematic diagram of a pedestrian re-identification system provided in an embodiment of the present invention;

[0062] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0063] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0064] This invention provides a pedestrian re-identification method. The execution entity of the pedestrian re-identification method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this invention: a server, a terminal, etc. In other words, the pedestrian re-identification method can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0065] Reference Figure 1 The diagram shown is a flowchart of a pedestrian re-identification method provided in an embodiment of the present invention. In this embodiment, the pedestrian re-identification method includes:

[0066] S1. Obtain pedestrian image data of the pedestrian to be identified, and preprocess the pedestrian image data to obtain a preprocessed image.

[0067] In this embodiment of the invention, the pedestrian image data of the pedestrian to be identified refers to the set of grayscale values ​​of each pixel of the pedestrian to be re-identified, which can be obtained by a data script, and the data script can be compiled by the JS scripting language.

[0068] Furthermore, in this embodiment of the invention, background noise can be removed by preprocessing the pedestrian image data to obtain background-free pedestrian images, which can then be used as the objects for subsequent operations. The preprocessing involves separating each text image and handing it over to the recognition module for identification. This includes processing performed before feature extraction, segmentation, and matching of the input image.

[0069] Furthermore, as an embodiment of the present invention, the preprocessing of the pedestrian image data to obtain a preprocessed image can be achieved by pixel-level segmentation of the image data using a pre-trained instance segmentation network model. Instance segmentation is a combination of object detection and semantic segmentation; it is a computer vision technique that detects objects in the input image and assigns a category label to each pixel of the object, enabling the differentiation of different instances with the same semantic category in the foreground.

[0070] S2. Local feature extraction is performed on the preprocessed image to obtain the local pedestrian features of the pedestrian to be identified.

[0071] In this embodiment of the invention, local features of the pedestrian after background noise removal can be obtained by performing local feature extraction on the preprocessed image.

[0072] Furthermore, as an embodiment of the present invention, the step of extracting local features from the preprocessed image to obtain the pedestrian local features of the pedestrian to be identified includes: identifying the structural position of the pedestrian's body in the preprocessed image; performing region segmentation on the preprocessed image according to the structural position to obtain segmented regions; extracting the regional features of the segmented regions and stitching the regional features together to obtain the stitched local features of the preprocessed image, and using the stitched local features as the pedestrian local features.

[0073] Furthermore, as an embodiment of the present invention, the identification of the structural location of the human body in the preprocessed image can be achieved through a human body parsing algorithm, such as identifying the locations of the head, body, left arm, right arm, left leg, and right leg in the preprocessed image. Here, human body parsing is a fine-grained semantic segmentation task, which aims to identify pixel-level components of a human image, such as body parts and clothing.

[0074] Furthermore, as an embodiment of the present invention, the step of segmenting the preprocessed image according to the structural position to obtain a segmented region includes: constructing a bounding rectangle of the structural position; and determining the segmented region of the preprocessed image according to the bounding rectangle.

[0075] Optionally, the extraction of regional features of the segmented region can be achieved using a convolutional neural network (CNN). The CNN is a type of feedforward neural network that includes convolutional computation and has a deep structure; it is one of the representative algorithms of deep learning.

[0076] Furthermore, as an embodiment of the present invention, the process of stitching together the regional features to obtain the stitched local features of the preprocessed image can be achieved through a stitching algorithm.

[0077] S3. Perform global feature extraction on the preprocessed image to obtain the global pedestrian features of the pedestrian to be identified;

[0078] In this embodiment of the invention, global features representing the overall information of pedestrians can be obtained by performing global feature extraction on the preprocessed image.

[0079] Furthermore, as an embodiment of the present invention, the global feature extraction of the preprocessed image to obtain the global pedestrian features of the pedestrian to be identified can be extracted by methods such as principal component analysis, linear discriminant analysis, and convolutional neural networks.

[0080] S4. Perform adaptive noise matching processing on the pedestrian local features and the pedestrian global features respectively to obtain adaptive local features and adaptive global features, and transform the adaptive local features and the adaptive global features into adaptive local vectors and adaptive global vectors.

[0081] The embodiments of the present invention can improve the robustness of re-identification by performing adaptive noise matching processing on the local features and global features of the pedestrian respectively, thereby making the local features and global features more differentiated and discriminative.

[0082] Further, as an embodiment of the present invention, the step of performing adaptive noise matching processing on the pedestrian local features and the pedestrian global features to obtain adaptive local features and adaptive global features includes: acquiring the segmented region corresponding to the pedestrian local features and the preprocessed image corresponding to the pedestrian global features; calculating the adaptive noise ratio of the segmented region to obtain the segmented region noise ratio, and calculating the adaptive noise ratio of the preprocessed image to obtain the global image noise ratio; performing adaptive noise processing on the segmented region and the preprocessed image respectively based on the segmented region noise ratio and the global image noise ratio to obtain a noisy segmented image and a noisy global image; and extracting features from the noisy segmented image and the noisy global image to obtain adaptive local features and adaptive global features.

[0083] Further, as an embodiment of the present invention, calculating the adaptive noise ratio of the segmented region and the preprocessed image includes: calculating the adaptive noise ratio of the segmented region using the following formula to obtain the segmented region noise ratio:

[0084]

[0085] Where β represents the segmentation region noise ratio, i represents the segmentation region of the pedestrian, j is a segmentation region different from i, k represents the number of matching layer weights in the adaptive noise matching network, l is the number of hidden layers, f(·) is the activation function, and x i y represents the input of the i-th segmentation region in the adaptive noise matching network. i Let W represent the output of the i-th segmentation region in the adaptive noise matching network, W be the weight parameter, λ be the regularization penalty coefficient, α be the adaptive noise adjustment factor, and θ be the value of the segmentation. i Let N be the adaptive noise matrix, and let N represent the total number of segmented regions for the same pedestrian.

[0086] Furthermore, as an embodiment of the present invention, the transformation of the adaptive local features and the adaptive global features into adaptive local vectors and adaptive global vectors can reduce the dimensionality of the adaptive local features and the adaptive global features to improve computational efficiency, and can be achieved through a vector transformation algorithm.

[0087] S5. Detect human key points in the preprocessed image and construct connecting edges between any of the human key points. Based on the human key points, construct the pose node matrix of the pedestrian to be identified and the pose edge matrix of the pedestrian to be identified based on the connecting edges.

[0088] This invention can capture important feature points of the human body structure to characterize local human features by detecting key human body points in the preprocessed image.

[0089] Furthermore, as an embodiment of the present invention, the detection of human key points in the preprocessed image includes: using a pre-trained human key point detection model to detect key points in the preprocessed image to obtain detected key points; marking the detected key points to obtain marking numbers; and generating the human key points based on the detected key points and the marking numbers.

[0090] Furthermore, in this embodiment of the invention, connections between key points can be established by constructing connecting edges between any of the aforementioned key points of the human body.

[0091] Furthermore, as an embodiment of the present invention, the construction of the connection edge between any of the human body key points can be obtained by connecting any two of the human body key points.

[0092] Furthermore, in this embodiment of the invention, by constructing the posture node matrix and posture edge matrix of the pedestrian to be identified based on the human body key points and the connecting edges, the posture patterns of pedestrians in the image, such as body shape and movement, can be mined as the posture features of the pedestrian to be identified.

[0093] Furthermore, as an embodiment of the present invention, the step of constructing the pose node matrix of the pedestrian to be identified based on the human body key points includes: extracting node features of the human body key points; constructing node feature vectors of the human body key points based on the node features and the label numbers of the human body key points; and vertically concatenating the node feature vectors according to the ascending order of the label numbers of the human body key points to obtain the pose node matrix.

[0094] Furthermore, as an embodiment of the present invention, constructing the pose edge matrix of the pedestrian to be identified based on the connecting edge includes: identifying the marker number of the human key point corresponding to the connecting edge; and calculating the edge length of the connecting edge using the following formula.

[0095]

[0096] Where, d ij x represents the length of the edge connecting the two key points of the human body, corresponding to the i and j-th points.i x j Let x and y represent the x-coordinates of the i-th and j-th key points of the human body, respectively. i y j Let i and j represent the ordinates of the i-th and j-th human body key points, respectively, where i and j represent the marker numbers of the human body key points, and k represents the number of human body key points.

[0097] The attitude edge matrix is ​​constructed based on the marker number and the edge length.

[0098] S6. Update and aggregate the attitude node matrix and the attitude edge matrix to obtain an aggregated node matrix and an aggregated edge matrix. Perform dimensionality reduction and concatenation on the aggregated node matrix and the aggregated edge matrix to obtain an aggregated attitude vector.

[0099] The embodiments of the present invention can enhance the pedestrian's posture features by updating and aggregating the posture node matrix and the posture edge matrix, thereby obtaining an enhanced posture feature representation.

[0100] Furthermore, as an embodiment of the present invention, the step of updating and aggregating the attitude node matrix and the attitude edge matrix to obtain the aggregated node matrix and the aggregated edge matrix includes: calculating the updated node matrix of the attitude node matrix using the following formula;

[0101]

[0102]

[0103]

[0104] Among them, M (l+1) Let σ represent the aggregation node matrix, σ represent the activation function, I be the identity matrix, W be the learned weight parameters, and l be the number of convolutional layers. Let S be the degree matrix of the edge characteristic matrix S. The elements on the main diagonal of the matrix represent the number of edges at each node;

[0105] Calculate the attitude edge matrix of the updated node matrix and use the attitude edge matrix as the updated edge matrix; when the updated node matrix satisfies a preset condition, use the updated node matrix as the aggregated node matrix and use the updated edge matrix corresponding to the updated node matrix as the aggregated edge matrix.

[0106] Furthermore, in this embodiment of the invention, the computational efficiency of pedestrian pose features is improved by performing dimensionality reduction and concatenation processing on the aggregated node matrix and the aggregated edge matrix.

[0107] Furthermore, as an embodiment of the present invention, the dimensionality reduction and concatenation processing of the aggregated node matrix and the aggregated edge matrix to obtain the aggregated attitude vector can be achieved by pooling algorithm and concatenation algorithm.

[0108] S7. The adaptive local vector, the adaptive global vector, and the aggregated pose vector are fused to obtain the pedestrian fusion vector. The cosine similarity between the pedestrian fusion vector and the pedestrian feature vector in the pre-built pedestrian database is calculated. Based on the cosine similarity, the recognition result of the pedestrian to be identified is determined.

[0109] The embodiments of the present invention can obtain more comprehensive and multi-dimensional pedestrian features of the pedestrian to be identified by fusing the adaptive local vector, the adaptive global vector and the aggregated pose vector.

[0110] Furthermore, as an embodiment of the present invention, the pedestrian fusion vector obtained by fusing the adaptive local vector, the adaptive global vector, and the aggregated pose vector can be calculated using the following formula:

[0111] X M =αX L +βX G +γX P

[0112] Among them, X M Let X represent the pedestrian fusion vector. L Let X represent the adaptive local vector. G X represents the adaptive global vector. P Let α, β, and γ represent the aggregated attitude vector, where α, β, and γ represent the weight coefficients, and α + β + γ = 1.

[0113] In this embodiment of the invention, the cosine similarity between the pedestrian fusion vector and the pedestrian feature vector in the pre-built pedestrian database can be used to determine the degree of similarity between the pedestrian to be identified and the pedestrians in the pre-built pedestrian database.

[0114] Furthermore, as an embodiment of the present invention, the calculation of the cosine similarity between the pedestrian fusion vector and the pedestrian feature vector in the pre-built pedestrian database can be obtained by a cosine similarity algorithm. The cosine similarity is evaluated by calculating the cosine of the angle between the two vectors.

[0115] Furthermore, in this embodiment of the invention, the final pedestrian re-identification can be achieved by determining the identification result of the pedestrian to be identified based on the cosine similarity.

[0116] Furthermore, as an embodiment of the present invention, determining the identification result of the pedestrian to be identified based on the cosine similarity includes: arranging the cosine similarities in descending order to obtain the descending sequence number of the cosine similarity; and determining the identification result of the pedestrian to be identified based on the descending sequence number.

[0117] As can be seen, this embodiment of the invention acquires pedestrian image data of the pedestrian to be identified, preprocesses the pedestrian image data to obtain a preprocessed image that removes background noise, resulting in a background-free pedestrian image. This preprocessed image serves as the object for subsequent operations. Local feature extraction on the preprocessed image yields local features of the pedestrian after background noise removal, while global feature extraction on the preprocessed image yields global features representing the overall information of the pedestrian. Furthermore, this embodiment of the invention, by performing adaptive noise matching processing on the pedestrian's local and global features respectively, obtains more differentiated and discriminative local and global features, thereby improving the robustness of re-identification. Converting the adaptive local and global features into adaptive local and global vectors reduces the dimensionality of the adaptive local and global features, improving computational efficiency. Detecting key human body points in the preprocessed image allows for the extraction of important features of the human body structure. By characterizing local human features and constructing connection edges between any of the human body's key points, relationships can be established between these key points. Based on the human body's key points and the connection edges, constructing the pose node matrix and pose edge matrix of the pedestrian to be identified can uncover the posture patterns such as body shape and movement of the pedestrian in the image, serving as the pose features of the pedestrian to be identified. Furthermore, in this embodiment of the invention, updating and aggregating the pose node matrix and the pose edge matrix can enhance the pedestrian's pose features, resulting in an enhanced pose feature expression. Fusing the adaptive local vector, the adaptive global vector, and the aggregated pose vector can yield more comprehensive and multi-dimensional pedestrian features of the pedestrian to be identified. Calculating the cosine similarity between the fused pedestrian vector and the pedestrian feature vector in the pre-constructed pedestrian database can determine the degree of similarity between the pedestrian to be identified and the pedestrians in the pre-constructed pedestrian database. Based on the cosine similarity, determining the identification result of the pedestrian to be identified can achieve the final pedestrian re-identification. Therefore, the pedestrian re-identification method and system proposed in this embodiment of the invention can improve the accuracy of pedestrian re-identification.

[0118] like Figure 2 The diagram shown is a functional block diagram of the pedestrian re-identification system of the present invention.

[0119] The pedestrian re-identification system 100 of this invention can be installed in an electronic device. Depending on the functions implemented, the pedestrian re-identification system may include an image preprocessing module 101, a local feature generation module 102, a global feature generation module 103, an adaptive noise matching module 104, a pose matrix generation module 105, an aggregated pose vector generation module 106, and a recognition result determination module 107. The modules described in this invention can also be referred to as units, which are a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and are stored in the memory of the electronic device.

[0120] In this embodiment, the functions of each module / unit are as follows:

[0121] The image preprocessing module 101 is used to acquire pedestrian image data of the pedestrian to be identified, and to preprocess the pedestrian image data to obtain a preprocessed image;

[0122] The local feature generation module 102 is used to extract local features from the preprocessed image to obtain the local features of the pedestrian to be identified.

[0123] The global feature generation module 103 is used to extract global features from the preprocessed image to obtain the global features of the pedestrian to be identified.

[0124] The adaptive noise matching module 104 is used to perform adaptive noise matching processing on the pedestrian local features and the pedestrian global features respectively to obtain adaptive local features and adaptive global features, and to convert the adaptive local features and the adaptive global features into adaptive local vectors and adaptive global vectors.

[0125] The pose matrix generation module 105 is used to detect human key points in the preprocessed image, construct connection edges between any human key points, construct the pose node matrix of the pedestrian to be identified based on the human key points, and construct the pose edge matrix of the pedestrian to be identified based on the connection edges.

[0126] The aggregated pose vector generation module 106 is used to update and aggregate the node feature matrix and the edge feature matrix to obtain an aggregated node matrix and an aggregated edge matrix, and to perform dimensionality reduction and concatenation processing on the aggregated node matrix and the aggregated edge matrix to obtain an aggregated pose vector.

[0127] The recognition result determination module 107 is used to fuse the adaptive local vector, the adaptive global vector, and the aggregated pose vector to obtain a pedestrian fusion vector, calculate the cosine similarity between the pedestrian fusion vector and the pedestrian feature vector in the pre-built pedestrian database, and determine the recognition result of the pedestrian to be identified based on the cosine similarity.

[0128] In detail, the modules in the pedestrian re-identification system 100 described in this embodiment of the invention employ the same methods as described above. Figure 1 The method used is the same as the pedestrian re-identification method described above and can produce the same technical effect, so it will not be repeated here.

[0129] It should be understood that the embodiments described are for illustrative purposes only and are not limited to this structure in terms of the scope of the patent invention.

[0130] In the several embodiments provided by this invention, it should be understood that the disclosed devices, systems, and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0131] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0132] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0133] It will be apparent 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 characteristics of the present invention.

[0134] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the invention. No appended diagram markings in the claims should be construed as limiting the scope of the claims.

[0135] The embodiments of this invention can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that utilize digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0136] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in a system claim may also be implemented by a single unit or device through software or hardware. The term "second class" is used to indicate names and does not indicate any specific order.

[0137] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A pedestrian re-identification method, characterized in that, The method includes: Obtain pedestrian image data of the pedestrian to be identified, and preprocess the pedestrian image data to obtain a preprocessed image; Local feature extraction is performed on the preprocessed image to obtain the local pedestrian features of the pedestrian to be identified; Global feature extraction is performed on the preprocessed image to obtain the global pedestrian features of the pedestrian to be identified; Adaptive noise matching processing is performed on the pedestrian local features and the pedestrian global features respectively to obtain adaptive local features and adaptive global features, and the adaptive local features and the adaptive global features are transformed into adaptive local vectors and adaptive global vectors; Detect human key points in the preprocessed image and construct connecting edges between any of the human key points. Based on the human key points, construct the pose node matrix of the pedestrian to be identified and the pose edge matrix of the pedestrian to be identified based on the connecting edges. The attitude node matrix and the attitude edge matrix are updated and aggregated to obtain an aggregated node matrix and an aggregated edge matrix. The aggregated node matrix and the aggregated edge matrix are then reduced in dimension and concatenated to obtain an aggregated attitude vector. By fusing the adaptive local vector, the adaptive global vector, and the aggregated pose vector, a pedestrian fusion vector is obtained. The cosine similarity between the pedestrian fusion vector and the pedestrian feature vector in the pre-built pedestrian database is calculated. Based on the cosine similarity, the recognition result of the pedestrian to be identified is determined.

2. The pedestrian re-identification method as described in claim 1, characterized in that, The step of extracting local features from the preprocessed image to obtain the local pedestrian features of the pedestrian to be identified includes: Identify the structural locations of pedestrian bodies in the preprocessed image; Based on the structural location, the preprocessed image is segmented to obtain segmented regions; Extract the regional features of the segmented region and stitch the regional features together to obtain the stitched local features of the preprocessed image. Use the stitched local features as the pedestrian local features.

3. The pedestrian re-identification method as described in claim 2, characterized in that, The step of segmenting the preprocessed image into segmented regions based on the structural position includes: Construct the bounding rectangle of the structure location; The segmentation region of the preprocessed image is determined based on the circumscribed rectangle.

4. The pedestrian re-identification method as described in claim 1, characterized in that, The step of performing adaptive noise matching processing on the pedestrian local features and the pedestrian global features respectively to obtain adaptive local features and adaptive global features includes: Obtain the segmented regions corresponding to the local features of the pedestrian and the preprocessed images corresponding to the global features of the pedestrian; Calculate the adaptive noise ratio of the segmented region to obtain the segmented region noise ratio, and calculate the adaptive noise ratio of the preprocessed image to obtain the global image noise ratio; Based on the noise ratio of the segmented region and the noise ratio of the global image, adaptive noise processing is performed on the segmented region and the preprocessed image respectively to obtain a noisy segmented image and a noisy global image. Feature extraction is performed on the noise segmentation image and the noise global image to obtain adaptive local features and adaptive global features.

5. The pedestrian re-identification method as described in claim 4, characterized in that, The calculation of the adaptive noise ratio of the segmented region to obtain the segmented region noise ratio includes: The adaptive noise ratio of the segmented region is calculated using the following formula to obtain the noise ratio of the segmented region: , in, Let represent the segmentation region noise ratio, where i represents the segmentation region of the pedestrian, j represents a segmentation region different from i, and k represents the number of matching layer weights in the adaptive noise matching network. To hide the number of layers, For activation function, This represents the input to the i-th segmentation region in the adaptive noise matching network. This represents the output of the i-th segmentation region in the adaptive noise matching network, where W is the weight parameter. Here, is the regularization penalty coefficient, and 'a' is the adaptive noise adjustment factor. Let N be the adaptive noise matrix, and let N represent the total number of segmented regions for the same pedestrian.

6. The pedestrian re-identification method as described in claim 1, characterized in that, The detection of human key points in the preprocessed image includes: The pre-trained human keypoint detection model is used to detect keypoints in the preprocessed image to obtain the detected keypoints. The key detection points are marked to obtain the marking sequence number; The human body key points are generated based on the detected key points and the marked sequence number.

7. The pedestrian re-identification method as described in claim 1, characterized in that, The step of constructing the pose node matrix of the pedestrian to be identified based on the key points of the human body includes: Extract the node features of the key points of the human body; Based on the node features and the labeling sequence number of the human body key points, construct the node feature vector of the human body key points; The node feature vectors are vertically concatenated according to the ascending order of the marker numbers of the human body key points to obtain the posture node matrix.

8. The pedestrian re-identification method as described in claim 1, characterized in that, The step of constructing the pose edge matrix of the pedestrian to be identified based on the connecting edges includes: Identify the marker numbers of the key human body points corresponding to the connecting edges; Calculate the length of the connecting edge using the following formula; ,in, This represents the length of the edge connecting the two key points of the human body, corresponding to the i-th and j-th points. , These represent the x-coordinates of the i-th and j-th key points of the human body, respectively. , These represent the ordinates of the i-th and j-th key points of the human body, respectively. These represent the marker numbers of key human body points, and k represents the number of key human body points. The attitude edge matrix is ​​constructed based on the marker number and the edge length.

9. The pedestrian re-identification method as described in claim 1, characterized in that, The step of updating and aggregating the attitude node matrix and the attitude edge matrix to obtain the aggregated node matrix and aggregated edge matrix includes: The updated node matrix of the attitude node matrix is ​​calculated using the following formula; , in, Represents an aggregate node matrix. This represents the activation function. Let W be the identity matrix and W be the learned weight parameters. The number of convolutional layers, Let S be the degree matrix of the edge characteristic matrix S. The elements on the main diagonal of the matrix represent the number of edges at each node; Calculate the attitude edge matrix of the updated node matrix, and use the attitude edge matrix as the updated edge matrix; When the updated node matrix meets the preset conditions, the updated node matrix is ​​used as the aggregated node matrix, and the updated edge matrix corresponding to the updated node matrix is ​​used as the aggregated edge matrix.

10. A pedestrian re-identification system, characterized in that, The system includes: The image preprocessing module is used to acquire pedestrian image data of the pedestrian to be identified, and to preprocess the pedestrian image data to obtain a preprocessed image; The local feature generation module is used to extract local features from the preprocessed image to obtain the local pedestrian features of the pedestrian to be identified. A global feature generation module is used to extract global features from the preprocessed image to obtain the global pedestrian features of the pedestrian to be identified. An adaptive noise matching module is used to perform adaptive noise matching processing on the pedestrian local features and the pedestrian global features respectively to obtain adaptive local features and adaptive global features, and to convert the adaptive local features and the adaptive global features into adaptive local vectors and adaptive global vectors. The pose matrix generation module is used to detect human key points in the preprocessed image, construct connection edges between any human key points, construct the pose node matrix of the pedestrian to be identified based on the human key points, and construct the pose edge matrix of the pedestrian to be identified based on the connection edges. An aggregated attitude vector generation module is used to update and aggregate the attitude node matrix and the attitude edge matrix to obtain an aggregated node matrix and an aggregated edge matrix, and to perform dimensionality reduction and concatenation on the aggregated node matrix and the aggregated edge matrix to obtain an aggregated attitude vector. The recognition result determination module is used to fuse the adaptive local vector, the adaptive global vector, and the aggregated pose vector to obtain a pedestrian fusion vector, calculate the cosine similarity between the pedestrian fusion vector and the pedestrian feature vector in the pre-built pedestrian database, and determine the recognition result of the pedestrian to be identified based on the cosine similarity.

Images