Gait recognition system based on artificial intelligence and recognition method thereof
The space-time graph convolutional neural network (STGCN) extracts energy images from gait video data and constructs space-time graphs, which solves the problems of large amount of calculation and incomplete feature extraction of traditional gait recognition methods, and realizes the complex environment High-precision gait recognition.
Patent Information
- Application Number
- CN202411107863.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-13
- Publication Date
- 2024-11-08
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Traditional gait recognition methods have problems such as large amount of computation, difficult to meet model assumptions, incomplete feature extraction, and difficult to obtain training data. In gait recognition, deep learning is difficult to effectively integrate spatio-temporal information and process dynamic changes, noise and occlusion.
Space-time graph convolutional neural network (STGCN) is used to extract energy images from gait video data and construct spatiotemporal maps. The changes in gait in space and time are captured through convolutional neural networks, and the classifier is used to achieve accurate identification.
It improves the accuracy and robustness of gait recognition, can effectively identify different gait patterns in complex environments, reduces the computational volume and model complexity, and enhances the application value.
Smart Images

Figure CN118918612A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of gait image recognition, and more specifically relates to a gait recognition system based on artificial intelligence and a recognition method thereof. Background Art
[0002] Gait recognition is also seen as a biometric identification method with great research potential. Compared with facial recognition and fingerprint recognition, gait recognition does not require the direct participation and cooperation of the user and can perform identity recognition without touch within a certain distance.
[0003] In traditional gait recognition, commonly used methods include model-based methods and motion-based methods. In the model-based method, gait data is obtained through a series of sensors, such as accelerometers and gyroscopes, or video images, and then a motion model describing the gait is constructed, such as a Markov model, a hidden Markov model, or a mixed Gaussian model. In the motion-based method, gait data can also be obtained through sensors or video images, and then recognized by extracting the gait's motion trajectory, or motion optical flow and other information.
[0004] However, these traditional gait recognition methods all have corresponding problems. First, a series of preprocessing steps are required, such as background elimination, foot positioning and gait division, which require a lot of calculations; second, the gait model needs to be based on a series of assumptions, such as walking on flat ground, constant speed, constant step length, etc., which are difficult to meet in actual environments; third, the extraction of gait features is based on artificial design, such as walking speed, walking step length, body swing, etc., which may not be able to fully describe all the information of gait; fourth, gait recognition requires a large amount of training data, which is difficult to obtain in actual applications.
[0005] As deep learning technology has made significant progress in image recognition and video processing, people have begun to try to apply it to gait recognition. However, there are still many problems in deep learning in gait recognition, such as how to effectively integrate the spatiotemporal information of gait, how to deal with the dynamic changes of gait, how to deal with video noise and occlusion, how to reduce the complexity and computational complexity of the model, etc. These problems need further research. Summary of the invention
[0006] The main technical problem addressed by the present invention is how to effectively integrate and process the spatiotemporal characteristics of gait to improve the accuracy and robustness of gait recognition. This mainly involves how to effectively extract gait characteristics from a large amount of gait video data, build a network model suitable for gait characteristics on this basis, and how to accurately recognize different gait patterns by training the network model.
[0007] In order to achieve the above object, the present invention is implemented by adopting the following technical scheme: the identification method comprises: Extract and construct gait energy image GEI from continuous gait video frames; Construct a space-time graph STG, which is a 3D graph in which two dimensions represent the spatial dimensions of the image and the other dimension represents time; Apply the space-time graph convolutional neural network STGCN to the space-time graph STG. The space-time graph convolutional neural network STGCN can capture the spatiotemporal dependencies of nodes in the graph; The output of the space-time graph convolutional neural network STGCN is classified by a classifier to obtain the result of gait recognition.
[0008] In one embodiment, constructing a gait energy image GEI comprises: Decompose the gait video sequence into consecutive frames; Detect and track pedestrians in video frames, i.e. object detection and target tracking; Identify and mark a complete walking cycle, and define the walking cycle using the contact time and swing time between footsteps; The captured walking cycle is binarized, with the pedestrian's gait outline represented as white and the background as black; All binary images in the walking cycle are averaged to generate the gait energy image GEI.
[0009] In one embodiment, constructing a spatiotemporal graph STG includes: Create an empty three-dimensional array to represent STG. The dimension of this array is (width, height, time), where width and height correspond to the width and height of the gait energy image GEI, and time corresponds to the number of frames in the gait cycle. Treat the three-dimensional array as a graph, where each pixel position is represented as a node; Traverse all gait energy images GEI and fill the pixel intensity value of each GEI into the space-time graph STG. The characteristics of each node (x, y, t) in the space-time graph STG are as follows: Let I(x,y,t) represent the intensity value corresponding to the position (x,y) in the gait energy image GEI at time t, and then regard the spatiotemporal graph STG as a three-dimensional function ,Right now, .
[0010] In one embodiment, the space-time graph convolutional neural network STGCN is applied to the space-time graph STG including: The input of the space-time graph convolutional neural network STGCN is the space-time graph STG of the gait energy image GEI; A spatial-temporal graph convolutional neural network (STGCN) is applied to each gait cycle, that is, each gait energy image GEI, to capture and learn the changes in space and time of gait, and extract in-depth features from the dynamic change graph of gait, namely the spatiotemporal graph STG.
[0011] In one embodiment, the output of the space-time graph convolutional neural network STGCN is classified, the high-dimensional features are converted into a form that can be classified, and the high-dimensional features are processed through a fully connected layer FC and a softmax function to obtain a gait recognition result; The fully connected layer converts the output of the spatial-temporal graph convolutional neural network STGCN into an intermediate feature representation through linear operations with the weight matrix and non-linear activation functions; a softmax function is used to assign a probability to each gait category, and the softmax function normalizes the output of the fully connected layer into a probability distribution.
[0012] On the other hand, a gait recognition system based on artificial intelligence is provided, wherein the system is applicable to the method described, and the system comprises: Gait energy image GEI building module, space-time graph STG building module, space-time graph convolutional neural network STGCN module and classifier module; A gait energy image GEI construction module is used to extract and construct a gait energy image GEI; The space-time graph STG construction module is used to construct a space-time graph STG, create an empty three-dimensional array to represent STG, and fill the pixel intensity values of the gait energy image GEI into the space-time graph STG; The space-time graph convolutional neural network STGCN module is used to apply the space-time graph STGCN to the space-time graph STG. The space-time graph convolutional neural network STGCN can capture the spatiotemporal dependencies of nodes in the graph; The classifier module is used to classify the output of the space-time graph convolutional neural network STGCN, convert the high-dimensional features into a form that can be classified, and process them through the fully connected layer and softmax function to obtain the result of gait recognition.
[0013] In one embodiment, the gait energy image GEI building module includes: an object detection and target tracking module for detecting and tracking pedestrians in video frames.
[0014] In one embodiment, the space-time graph convolutional neural network STGCN module includes: a network structure initialization module and a network training module. The network structure initialization module is used to initialize the network structure of STGCN, and the network training module is used to train the network model by learning the characteristics of the nodes in the space-time graph and how the characteristics change in space and time.
[0015] In one embodiment, the gait recognition system includes a display module for displaying the result of gait recognition to the user.
[0016] Beneficial effects of the present invention: The present invention proposes a novel gait recognition method and system based on deep learning. By using the space-time graph convolutional neural network STGCN, the spatial and temporal changes of gait can be effectively captured and learned, thereby realizing accurate recognition of different gait patterns. Compared with traditional gait recognition methods, the present invention has higher recognition accuracy and robustness, and can effectively perform gait recognition in various complex environments and conditions, such as noise interference, occlusion, etc., greatly improving the practicality and application value of gait recognition. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] Figure 1 A flow chart of an embodiment of the method of the present invention; Figure 2 A gait energy image flow chart is constructed for one embodiment of the method of the present invention; Figure 3 Construct a time-space diagram flow chart for an embodiment of the method of the present invention; Figure 4 A system diagram of an embodiment of the system of the present invention. DETAILED DESCRIPTION
[0018] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Typical embodiments of the present invention are given in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described in the present invention. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.
[0019] Unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those 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.
[0020] Example 1: Figure 1 As shown, the gait recognition method based on artificial intelligence includes: S1. Extract and construct gait energy images GEI from continuous gait video frames, including: like Figure 2 As shown, S101, decomposing a gait video sequence into continuous frames; First, you need to read the video file. You can use libraries in various programming languages to do this. For example, in Python, you can use the OpenCV library. In OpenCV, you can use the cv2.VideoCapture() function to open the video file.
[0021] Since the video needs to be decomposed into continuous frames, it is necessary to know the frame rate of the video, that is, how many frames per second. This can also be obtained through the cv2.VideoCapture().get(cv2.CAP_PROP_FPS) function in OpenCV.
[0022] After knowing the frame rate, the next step is to read and extract each frame of the video. In OpenCV, you can use the cv2.VideoCapture().read() function to read. In this function, two values are returned, one is a Boolean value indicating whether the read is successful, and the other is the frame read. After reading each frame, you can save it as an image file for subsequent processing. In OpenCV, you can use the cv2.imwrite() function to save.
[0023] S102: Detect and track pedestrians in the video frames, i.e., object detection and target tracking. The method used is the optical flow method. The optical flow method is a description of the movement speed and movement direction of pixels in a spatial sequence image. By calculating the movement changes of pixels between adjacent image frames, the local movement information of the pixels can be obtained, thereby realizing the detection and tracking of pedestrians.
[0024] S103, identifying and marking a complete walking cycle, and defining the walking cycle using the contact time and swing time between footsteps.
[0025] The frame difference method is used to detect foot movement. The frame difference method detects the moving object by calculating the pixel difference between the current frame and the previous frame. For the detection of the gait cycle, we can only focus on the movement of the foot area and use the frame difference method to detect the movement of the footsteps.
[0026] S104, binarizing the captured walking cycle, representing the pedestrian's gait outline as white and the background as black.
[0027] Convert the RGB three-channel color image to a grayscale image, because the binarization process does not require color information, only light and dark information. In order to convert the image into a binary image (background and object have only two colors, black and white), you need to set a threshold. All pixel values greater than this threshold will be set to one value (such as 255, representing white), and all pixel values less than this threshold will be set to another value (such as 0, representing black). The setting of the threshold depends on the color intensity difference between the object to be segmented and the background. Using the set threshold, binarize the grayscale image. This can be achieved by looping through each pixel value and comparing it with the threshold.
[0028] S105, average all the binary images in the walking cycle to generate a gait energy image GEI. Specifically, GEI is defined as the average value of the frames in a walking cycle. Assume that there are N frames in total, and the binary image of each frame is , represents the value of pixel (i, j) in the tth frame), then the calculation formula of GEI is: ; Where t=1,2,....,N.
[0029] This gait energy image (GEI) can capture the dynamic information of the walking cycle and encode it into a static image.
[0030] like Figure 3 As shown, S2, construct a spatiotemporal graph STG, which is a 3D graph in which two dimensions represent the spatial dimensions of the image and the other dimension represents time. In this spatiotemporal graph, each node represents a pixel position, and at each time step, the intensity of the pixel can be used as a node feature.
[0031] S201. Create an empty three-dimensional array to represent STG. The dimension of this array may be (width, height, time), where width and height correspond to the width and height of GEI, and time corresponds to the number of frames in the gait cycle. This three-dimensional array may be viewed as a graph, where each pixel position is represented as a node.
[0032] S202, traverse all GEIs and fill the pixel intensity value of each GEI into STG. Therefore, for each node (x, y, t) in STG, its characteristic can be defined as the intensity of the pixel position (x, y) in GEIt. Represents the characteristics of the node (x, y, t) in STG, then: make represents the intensity value corresponding to the position (x, y) in the gait energy image GEI at time t, and the space-time graph STG is regarded as a three-dimensional function Right now, ; This function defines the pixel intensity at position (x, y) and time t in physical space. This three-dimensional function F represents the so-called space-time graph. Among them, (x, y) is the two-dimensional spatial coordinate of the image, and t is the vertical coordinate corresponding to the time step. For each fixed t, In fact, it corresponds to GEI.
[0033] That is to say, continuous two-dimensional GEI images are superimposed to form a three-dimensional space-time graph (STG), in which the first two dimensions represent the spatial dimensions of the image and the third dimension is the time dimension.
[0034] In gait recognition, we hope to capture the dynamic information of a person while walking, so we need to integrate the time information (i.e., GEI image sequence) into the spatial information to form a three-dimensional structure of spatiotemporal data.
[0035] S203, inputting STG into a space-time graph convolutional neural network (STGCN) for training. The network can recognize different gait patterns by learning the characteristics of nodes in the space-time graph and how these characteristics change in space and time.
[0036] S3. Apply the space-time graph convolutional neural network STGCN to the space-time graph STG. For gait recognition, this means that STGCN can capture the changes in gait over time and the mutual influence of gait between different parts. The space-time graph STG contains both spatial (that is, the width and height of GEI) and temporal information (that is, the number of frames in the gait cycle).
[0037] A spatial-temporal graph convolutional neural network (STGCN) is used to process this information. STGCN is a special type of neural network that can process graph-structured data like STG. It can capture the spatial and temporal information of gait and recognize the important patterns in it. For the gait recognition task, this makes it possible to consider both the spatial structure of a single gait (i.e. the shape and relative position of each part of the body) and the temporal trend of the gait (i.e. the continuity and cyclicity of the gait action) at the same time. STGCN can learn how to distinguish the gait characteristics of different people from a gait cycle, which can be used for gait recognition.
[0038] The input of the space-time graph convolutional neural network STGCN is the space-time graph STG of the gait energy image GEI; STGCN consists of two main parts: spatial graph convolution layer and temporal convolution layer. The goal of the spatial graph convolution layer is to capture the dependency between a node and its neighboring nodes, that is, the spatial dependency in the graph. This is achieved by applying convolution operations on the nodes in the graph, where the specific convolution weights depend on the relative positions of the nodes. Mathematical formula: ; Here A is the adjacency matrix of the node, D is the diagonal matrix constructed with the sum of the rows of A (node degree), X is the node feature matrix, W is the weight parameter to be learned, and X' is the node feature matrix after the spatial convolution operation.
[0039] The goal of the temporal convolutional layer is to capture the dynamic changes of node features in consecutive time steps, that is, the temporal dependencies in the graph. This is achieved by applying a one-dimensional convolution operation on the node features of multiple time steps and adding a bias term.
[0040] ; Here conv1d represents a one-dimensional convolution operation, H is the node feature matrix of multiple time steps, W is the weight parameter to be learned, b is the bias parameter, and H' is the node feature matrix after the time convolution operation.
[0041] STGCN is applied on each gait cycle (each GEI image) to capture and learn the spatial and temporal changes of gait. In this way, richer and deeper features can be extracted from the dynamic change map of gait (STG image).
[0042] S4. Classify the output of the space-time graph convolutional neural network STGCN through a classifier to obtain the result of gait recognition.
[0043] After being processed by STGCN, a representation with rich spatiotemporal features is obtained. However, this representation cannot be directly used for classification tasks, and this high-dimensional feature needs to be converted into a form that can be classified. The result of gait recognition is obtained by processing through a fully connected layer (FC) and a softmax function.
[0044] The fully connected layer converts the output of the spatial-temporal graph convolutional neural network STGCN into an intermediate feature representation through linear operations with the weight matrix and nonlinear activation functions; assuming that the output of STGCN is H' and the weight matrix is , the bias is The calculation process of the fully connected layer can be described as: ; Here, * represents the matrix multiplication operation.
[0045] The weight matrix W_fc is randomly initialized before network training begins. During the training process, the model parameters (including the weight matrix W_fc) are optimized through back-propagation and optimization algorithms. , bias ) to update.
[0046] The specific process of updating parameters includes: Forward propagation: Based on the current model parameters, the prediction results are calculated and the loss function (such as cross entropy) is calculated.
[0047] Backpropagation: Calculate the partial derivatives (gradients) of the loss function with respect to each model parameter.
[0048] Update parameters: Update the model parameters based on the gradient and a predefined learning rate.
[0049] For the weight matrix The specific update formula of , assuming that the learning rate is η and the loss function is L, then the update step can be performed according to the following formula: ; Using ReLU as the nonlinear activation function, the output of the fully connected layer can be expressed as: ; Where ReLU(x)=max(0,x).
[0050] The softmax function is used to assign a probability to each gait category. Suppose the output of the fully connected layer is A. The fully connected layer extracts features useful for identification from the spatiotemporal features of the gait. Then the softmax function normalizes each gait into a probability distribution.
[0051] In order to obtain a probability output that can be used for classification, a softmax function is used to assign a probability to each gait category. Let the output of the fully connected layer be A, then the softmax function can normalize it to a probability distribution: ; in, represents the probability of the i-th gait category, and is the component of the fully connected layer output, and Σ represents the sum of all possible gait categories.
[0052] In this way, the output of the softmax function is the probability distribution of the gait categories predicted by the model, and the category with the largest probability is taken as the prediction result of the model.
[0053] Example 2: Figure 4As shown, a gait recognition system based on artificial intelligence is applicable to the method described above, and the system comprises: Gait energy image GEI building module, space-time graph STG building module, space-time graph convolutional neural network STGCN module and classifier module.
[0054] Gait energy image GEI construction module is used to extract and construct gait energy image GEI; the gait energy image GEI construction module is a module used to extract and construct gait energy image GEI from continuous gait video frames. Its main function is to average all binary images in the walking cycle to generate gait energy image GEI.
[0055] The gait energy image GEI building module includes: an object detection and target tracking module, which is used to detect and track pedestrians in video frames.
[0056] The space-time graph STG construction module is used to construct a space-time graph STG, create an empty three-dimensional array to represent STG, and fill the pixel intensity values of the gait energy image GEI into the space-time graph STG; it realizes the construction of the space-time graph by creating an empty three-dimensional array to represent STG and filling the pixel intensity values of the gait energy image GEI into the space-time graph STG.
[0057] The space-time graph convolutional neural network STGCN module is used to apply the space-time graph STGCN to the space-time graph STG. The space-time graph convolutional neural network STGCN can capture the space-time dependencies of the nodes in the graph. The space-time graph convolutional neural network STGCN module includes: a network structure initialization module and a network training module, and is used to train the network model by learning the characteristics of the nodes in the space-time graph and how the characteristics change in space and time. The network structure initialization module is mainly used to initialize the network structure of STGCN, while the network training module improves the performance of the network model through learning and training.
[0058] The classifier module is used to classify the output of the space-time graph convolutional neural network STGCN, convert the high-dimensional features into a form that can be classified, and process them through a fully connected layer and a softmax function to obtain the result of gait recognition. The module is used to classify the output of the space-time graph convolutional neural network STGCN, convert the high-dimensional features into a form that can be classified, and process the output of STGCN through a fully connected layer and a softmax function to obtain the result of gait recognition. The gait recognition system includes a display module for displaying the result of gait recognition to the user. This module is mainly used to display the result of gait recognition in a visual way so that the user can intuitively understand the result of gait recognition.
[0059] Embodiment 3: First, gait video data is obtained, the video data is split into frames, the outline of the walking person is found, and the captured walking cycle is binarized, and the gait outline of the pedestrian is represented as white and the background is black. For example, for a 6-second video, the frame rate is 30fps, so a total of 180 frames of video images can be obtained.
[0060] Then, through the gait energy image GEI construction module, all the binary images in each walking cycle are averaged to generate the gait energy image GEI. Assuming that the walking cycle is 0.6 seconds, each walking cycle is 18 frames, so a total of 10 gait energy images can be obtained.
[0061] Next, a spatiotemporal graph is constructed through the spatiotemporal graph STG construction module. If the size of the gait energy image is WxH (e.g., 64x128), the size of the created spatiotemporal graph is WxHxT (e.g., 64x128x18).
[0062] Afterwards, the space-time graph STGCN is applied to the space-time graph STG through the space-time graph convolutional neural network STGCN module, for example, an 8-layer STGCN structure is used so that the characteristics in the space-time graph can be fully captured.
[0063] Finally, the output of STGCN is classified through the classifier module. For example, if there are 10 people, the softmax function is used to normalize each gait into 10 probability distributions, and finally the probability distribution of each frame gait corresponding to 10 people is obtained.
[0064] By tracking the gait with the largest probability distribution in the gait cycle, we can find out who the pedestrian corresponds to in real time.
[0065] The entire process can be carried out in real time and can run on various devices such as computers, servers, mobile devices, etc. It can be used for identity recognition in various fields, such as entry and exit management of communities, schools, and companies, and ultra-early warning of diseases.
[0066] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM).
[0067] It should be understood that the detailed description of the technical solutions of the present invention by means of the preferred embodiments is illustrative rather than restrictive. A person skilled in the art may modify the technical solutions described in the embodiments, or replace some of the technical features by equivalents, based on reading the specification of the present invention; and these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A gait recognition method based on artificial intelligence, characterized in that: The identification method comprises: Extract and construct gait energy image GEI from continuous gait video frames; Construct a space-time graph STG, which is a three-dimensional graph, in which two dimensions represent the spatial dimensions of the image and the other dimension represents time; Apply the space-time graph convolutional neural network STGCN to the space-time graph STG. The space-time graph convolutional neural network STGCN is used to capture the spatiotemporal dependencies of nodes in the graph. The output of the space-time graph convolutional neural network STGCN is classified by a classifier to obtain the result of gait recognition.
2. The artificial intelligence-based gait recognition method according to claim 1, characterized in that: The step of constructing the gait energy image GEI comprises: Decompose the gait video sequence into consecutive frames; Detect pedestrians in video frames and track them; Identify and mark a complete walking cycle; The captured walking cycle is binarized, with the pedestrian's gait outline represented as white and the background as black; All binary images in the walking cycle are averaged to generate the gait energy image GEI.
3. The gait recognition method based on artificial intelligence according to claim 1, characterized in that: The construction of a space-time graph STG includes: Create an empty three-dimensional array width, height, time to represent STG, where width and height correspond to the width and height of the gait energy image GEI, and time corresponds to the number of frames in the gait cycle. Treat the three-dimensional array as a graph in which each pixel position is represented as a node; Traverse all gait energy images GEI and fill the pixel intensity value of each GEI into the space-time graph STG. The characteristics of each node (x, y, t) in the space-time graph STG are as follows: Let I(x,y,t) represent the intensity value corresponding to the position (x,y) in the gait energy image GEI at time t, and then regard the spatiotemporal graph STG as a three-dimensional function ,Right now, 。 4. The gait recognition method based on artificial intelligence according to claim 1, characterized in that: The space-time graph convolutional neural network STGCN is applied to the space-time graph STG including: The input of the space-time graph convolutional neural network STGCN is the space-time graph STG of the gait energy image GEI; A spatial-temporal graph convolutional neural network (STGCN) is applied to each gait energy image GEI to capture and learn the spatial and temporal changes of gait and extract in-depth features from the spatiotemporal graph STG.
5. The gait recognition method based on artificial intelligence according to claim 1, characterized in that: The output of the space-time graph convolutional neural network STGCN is classified, the high-dimensional features are converted into a form that can be classified, and the high-dimensional features are processed through the fully connected layer FC and the softmax function to obtain the result of gait recognition; The fully connected layer converts the output of the spatial-temporal graph convolutional neural network STGCN into an intermediate feature representation through linear operations with the weight matrix and nonlinear activation functions; A softmax function is used to assign a probability to each gait category, and the softmax function normalizes the output of the fully connected layer into a probability distribution.
6. A gait recognition system based on artificial intelligence, wherein the system is applicable to the method according to any one of claims 1 to 5, characterized in that: The system comprises: Gait energy image GEI building module, space-time graph STG building module, space-time graph convolutional neural network STGCN module and classifier module; A gait energy image GEI construction module is used to extract and construct a gait energy image GEI; The space-time graph STG construction module is used to construct a space-time graph STG, create an empty three-dimensional array to represent STG, and fill the pixel intensity values of the gait energy image GEI into the space-time graph STG; The space-time graph convolutional neural network STGCN module is used to apply the space-time graph STGCN to the space-time graph STG. The space-time graph convolutional neural network STGCN is used to capture the spatiotemporal dependencies of nodes in the graph; The classifier module is used to classify the output of the space-time graph convolutional neural network STGCN, convert the high-dimensional features into a form that can be classified, and process them through the fully connected layer and softmax function to obtain the result of gait recognition.
7. The artificial intelligence-based gait recognition system according to claim 6, characterized in that: The gait energy image GEI building module includes: an object detection and target tracking module, which is used to detect and track pedestrians in video frames.
8. The artificial intelligence-based gait recognition system according to claim 6, characterized in that: The space-time graph convolutional neural network STGCN module includes: a network structure initialization module and a network training module. The network structure initialization module is used to initialize the network structure of STGCN, and the network training module is used to train the network model by learning the characteristics of the nodes in the space-time graph and how the characteristics change in space and time.
9. The artificial intelligence-based gait recognition system according to claim 6, characterized in that: The gait recognition system comprises a display module for displaying the result of gait recognition.
