Personnel non-inductive positioning method and system based on gait recognition

Abstract

According to the gait recognition-based personnel non-inductive positioning method and system provided by the invention, the gait feature vector with space-time relevance is constructed through multi-modal data collaborative acquisition of a visible light imaging device and a millimeter wave radar and in combination with a three-dimensional reconstruction technology and a fusion mechanism; the constraint relation of skeleton nodes is utilized to eliminate individual body type difference interference, and finally a high-robustness gait recognition model is formed; the system improves the real-time positioning precision of personnel, can penetrate through the interference of smog / rain and snow environments, effectively solves the problem of identity misjudgment in a shielding scene, and provides all-weather and non-contact accurate personnel tracking and behavior analysis capabilities for scenes such as intelligent security and public management.

Description

Technical Field

[0001] The present invention relates to the field of management and monitoring of industrial production safety, and in particular to a personnel non-sensing positioning method and system based on gait recognition. Background Art

[0002] Currently, the requirements for safety management in the industrial and production fields are increasing day by day, and the management and monitoring of equipment and personnel have become an important guarantee for safe production.

[0003] However, traditional equipment monitoring relies on manual record-keeping, which is subjective and time-consuming, making objective, real-time evidence collection difficult. Personnel monitoring relies on electronic tags or fixed sensor networks, but video surveillance and biometrics devices offer incomplete coverage and insufficient dynamic monitoring capabilities, making intelligent monitoring and early warning difficult. Furthermore, with the increasing complexity of production scenarios, a single data source can no longer fully reflect actual on-site conditions, necessitating an urgent need to improve safety risk identification and hidden danger prediction capabilities.

[0004] The rapid development of big data, artificial intelligence (AI) and large language model technologies, and the gradual maturity of cross-modal perception and cognitive processing capabilities, have provided a new technical path for improving security monitoring and control levels.

[0005] This paper proposes a personnel positioning and behavior monitoring method based on gait recognition and combined with video sequence analysis, which can achieve accurate individual tracking without electronic tags, effectively deal with environmental interference and individual differences, and provide reliable data support for on-site security. Summary of the Invention

[0006] In a first aspect of the present disclosure, a method for detecting a person's position without sensing the person's position based on gait recognition is provided, comprising the following steps:

[0007] The two-dimensional video stream data of the target area is collected by a visible light imaging device, and the point cloud data of the target area is collected by a millimeter wave radar;

[0008] Performing three-dimensional reconstruction on the two-dimensional video stream data to generate a temporal feature sequence including spatial coordinates;

[0009] Inputting the temporal feature sequence and the point cloud data into a pre-trained gait recognition model to extract an identity feature vector;

[0010] Multi-target trajectory association is achieved based on the identity identification feature vector matching, and the positioning coordinates of the personnel and the behavior classification results of the personnel are output in combination with the behavior semantics.

[0011] In combination with the first aspect, performing three-dimensional reconstruction on the two-dimensional video stream data to generate a temporal feature sequence including spatial coordinates includes:

[0012] Extracting temporally continuous video frames from the two-dimensional video stream data;

[0013] Performing three-dimensional dynamic reconstruction based on the video frames to generate model data with spatial coordinates;

[0014] Data enhancement processing is performed on the model data to generate a temporal feature sequence under different environmental interferences, where the different environments include rain, snow, and smoke.

[0015] In combination with the first aspect, the method for constructing the pre-training model includes:

[0016] Construct the first feature extractor including spatiotemporal convolutional network and Transformer structure;

[0017] Adding the constraint relationship of human skeleton nodes to the first feature extractor to obtain a second feature extractor;

[0018] Distributed model training is performed on the second feature extractor to obtain a pre-trained model, and the nodes of the second feature extractor update the model parameters based on the differential privacy mechanism.

[0019] In combination with the first aspect, inputting the temporal feature sequence and the point cloud data into a pre-trained gait recognition model to extract an identity feature vector includes:

[0020] Converting the time series feature sequence into a pulse interval time sequence and accumulating pulse signals;

[0021] Establishing a mapping matrix between the time series feature sequence and the point cloud data according to the pulse signal accumulation, and fusing the time series feature sequence with the point cloud data to obtain a fusion feature;

[0022] The fusion features are discretized to generate a feature codebook, an identity identification feature vector of the feature codebook is extracted, and the feature codebook is updated by a clustering algorithm.

[0023] In combination with the first aspect, the multi-target trajectory association based on the identity identification feature vector matching includes calculating the similarity of target features between adjacent frames in the identity identification feature vector, and determining that the targets are the same when the similarity is greater than a preset threshold.

[0024] In conjunction with the first aspect, the outputting of individual positioning coordinates and behavior classification results in combination with behavior semantics includes:

[0025] Extracting the behavioral semantics of the same target and its displacement trajectory within the target area;

[0026] Obtaining the positioning coordinates of the person according to the displacement trajectory;

[0027] The behavior semantics are analyzed and combined with the positioning coordinates to define the behavior classification results of the person, the behavior classification results including a first classification and a second classification, the first classification including walking, running, and leaving items behind, and the second classification determining wandering behavior based on a stay time threshold.

[0028] A second aspect of the present disclosure provides a sensorless positioning system for personnel based on gait recognition, comprising:

[0029] A data acquisition unit collects two-dimensional video stream data of the target area through a visible light imaging device and collects point cloud data of the target area through a millimeter wave radar;

[0030] A feature generation unit, performing three-dimensional reconstruction on the two-dimensional video stream data to generate a temporal feature sequence including spatial coordinates;

[0031] A feature fusion unit, inputting the temporal feature sequence and the point cloud data into a pre-trained gait recognition model to extract an identity feature vector;

[0032] The trajectory behavior association unit realizes multi-target trajectory association based on the identity identification feature vector matching, and outputs the positioning coordinates of the personnel and the behavior classification results of the personnel in combination with the behavior semantics.

[0033] According to a third aspect of the present disclosure, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory, wherein when the processor executes the program, the method for seamless positioning of a person based on gait recognition is implemented.

[0034] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided, which stores computer program instructions, and when the instructions are executed by a processor, the method for seamless positioning of a person based on gait recognition is implemented.

[0035] Beneficial effects: The present disclosure provides a method and system for non-sensing positioning of personnel based on gait recognition. By collaboratively collecting multimodal data of visible light imaging devices and millimeter-wave radars, combined with three-dimensional reconstruction technology and fusion mechanisms, a gait feature vector with spatiotemporal correlation is constructed; the constraint relationship of skeletal nodes is used to eliminate the interference of individual body shape differences, and finally a highly robust gait recognition model is formed; the real-time positioning accuracy of personnel is improved, and the system can penetrate the interference of smoke / rain and snow environments, effectively solving the problem of identity misjudgment in obstructed scenes, and providing all-weather, non-contact, accurate personnel tracking and behavior analysis capabilities for scenarios such as smart security and public management. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is a flow chart of a method for sensorless positioning of a person based on gait recognition according to an embodiment of the present disclosure;

[0037] Figure 2Schematic diagram of the structure of a sensorless positioning system for personnel based on gait recognition according to an embodiment of the present disclosure;

[0038] Figure 3 Schematic diagram of an electronic device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0039] Exemplary embodiments are described in detail herein, with examples illustrated in the accompanying drawings. When the following description refers to the drawings, identical numerals in different drawings represent identical or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with the present disclosure.

[0040] The terms used in the embodiments of the present disclosure are for the purpose of describing specific embodiments only and are not intended to limit the embodiments of the present disclosure. The singular forms "a," "the," and "the" used in the embodiments of the present disclosure and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0041] It should be understood that although the terms first, second, third, etc. may be used to describe various information in the embodiments of the present disclosure, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of the embodiments of the present disclosure, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0042] like Figure 1 FIG. 1 is a flow chart of a method for sensorless positioning of a person based on gait recognition according to an embodiment of the present disclosure, including:

[0043] S101: Collecting two-dimensional video stream data of the target area through a visible light imaging device and collecting point cloud data of the target area through a millimeter wave radar;

[0044] S102: Performing three-dimensional reconstruction on the two-dimensional video stream data to generate a temporal feature sequence including spatial coordinates;

[0045] S103: Inputting the temporal feature sequence and the point cloud data into a pre-trained gait recognition model to extract an identity feature vector;

[0046] S104: Implement multi-target trajectory association based on the identity identification feature vector matching, and output the positioning coordinates of the personnel and the behavior classification results of the personnel in combination with the behavior semantics.

[0047] For example, the system synchronously acquires sensor data of the target area through a heterogeneous sensor array. The visible light imaging device uses a high-resolution CMOS sensor (such as the Sony IMX686) to capture RGB video streams at a frame rate of 30fps. Its spectral response range covers the visible light band of 400-700nm, ensuring imaging quality under normal lighting conditions.

[0048] The millimeter-wave radar module utilizes 77GHz FMCW frequency-modulated continuous wave (FMCW) technology. Its MIMO antenna array achieves a scanning range of ±60° in azimuth and ±30° in elevation. The output point cloud data includes target range, velocity, and position. To achieve spatiotemporal alignment of multi-source data, the system incorporates hardware-level synchronization trigger signals. Using a GPS clock module, the system maintains a timing deviation of less than 1ms for each sensor. A spatiotemporal registration algorithm based on feature point matching is employed to map radar point cloud coordinates to the visual imaging plane, establishing a unified spatiotemporal reference system.

[0049] The system uses an improved Neural Radiance Field (NeRF) technique to perform 3D dynamic reconstruction of 2D video stream data. Specifically, continuous video frames are fed into a multi-layer perceptron (MLP) network, which uses a ray tracing algorithm to predict the density σ and color value c of the sampling points, generating a 3D voxel model with physical measurement properties.

[0050] At the same time, the system integrates a physical simulation engine (such as NVIDIA PhysX) to simulate the impact of the swing of clothing made of different materials and the change of ground friction coefficient on gait characteristics, generate an adversarial sample library containing interference from rain, snow, and smoke, and enhance the environmental adaptability of the model.

[0051] The pre-trained gait recognition model uses a three-level feature fusion architecture: the first feature extractor deploys a spiking neural network (SNN) to convert high-frame-rate video streams into pulse emission timing codes, and uses a biologically inspired integral emission model to capture microsecond-level motion differences;

[0052] The second feature extractor uses a graph convolutional network (GCN) to construct a dynamic topology map based on the millimeter-wave radar point cloud, modeling the coupling relationship between the micro-Doppler features of the target surface and the human skeleton nodes;

[0053] The resulting pre-trained model dynamically weights visual appearance features and radar motion features using a learnable weight matrix to generate a 128-dimensional identity feature vector. To improve feature discrimination, the system discretizes the continuous feature space using a vector quantization encoder (VQ-VAE), constructing a feature codebook containing 4096 codewords. A contrastive loss function is used to optimize both intra-class compactness and inter-class separability.

[0054] In the multi-target tracking process, the dual criteria of appearance feature similarity and motion trajectory continuity are integrated.

[0055] Exemplarily, the cosine similarity between the feature vector of the current frame and the historical cache is calculated, and identity association is triggered when the similarity threshold exceeds 0.85.

[0056] Optionally, the trajectory prediction model integrates the extended Kalman filter and the radar motion equation, and fuses the visual observation data with the original radar trajectory through the covariance cross algorithm, so as to maintain the trajectory continuity even when the target is briefly occluded (≤5 seconds).

[0057] Behavioral semantic analysis performs short-time Fourier transform and wavelet packet decomposition on limb movement signals, extracts energy distribution characteristics in the 0.5-5Hz frequency band, combines it with a predefined behavioral ontology knowledge base (containing 23 categories of abnormal behavioral semantic rules), and outputs the behavioral level through a fuzzy inference engine.

[0058] The system ultimately generates a structured data packet containing timestamps, geographic coordinates (WGS84 format), behavior tags, and confidence scores, and achieves real-time response within 200ms through edge computing nodes.

[0059] Furthermore, the three-dimensional reconstruction of the two-dimensional video stream data to generate a temporal feature sequence including spatial coordinates includes:

[0060] Extracting temporally continuous video frames from the two-dimensional video stream data;

[0061] Performing three-dimensional dynamic reconstruction based on the video frames to generate model data with spatial coordinates;

[0062] Data enhancement processing is performed on the model data to generate a temporal feature sequence under different environmental interferences, where the different environments include rain, snow, and smoke.

[0063] Specifically, a frame synchronization controller extracts sequential video frames, sampling at 30fps to ensure motion continuity. The 3D dynamic reconstruction process then performs ray casting calculations on the density and color of each pixel to optimize human pose estimation.

[0064] During the data enhancement phase, a physical simulation engine is deployed to simulate the raindrop collision dynamics model in rainy and snowy weather and the particle diffusion effect in a smoky environment, generating adversarial samples containing environmental interference.

[0065] The generation of temporal feature sequences is achieved through a spatiotemporal encoder, which converts the three-dimensional model data into a four-dimensional tensor containing spatial coordinates (X, Y, Z) and timestamps (T).

[0066] Furthermore, the method for constructing the pre-training model includes:

[0067] Construct the first feature extractor including spatiotemporal convolutional network and Transformer structure;

[0068] Adding the constraint relationship of human skeleton nodes to the first feature extractor to obtain a second feature extractor;

[0069] Distributed model training is performed on the second feature extractor to obtain a pre-trained model, and the nodes of the second feature extractor update the model parameters based on the differential privacy mechanism.

[0070] In the pre-training model construction method, the first feature extractor adopts a hybrid architecture of three-dimensional spatiotemporal convolutional network (3D-STCNN) and Transformer.

[0071] For example, the spatiotemporal convolutional network uses a 5×5×3 convolution kernel (corresponding to H×W×T dimensions) and performs transfer learning based on the ImageNet pre-trained weights to capture local motion patterns.

[0072] The Transformer module is equipped with an 8-head self-attention mechanism to model long-term dependencies across frames. The introduction of human skeletal node constraints involves extracting 17 key points based on the OpenPose framework to construct a graph-structured data structure. Kinematic constraint equations for adjacent joints (e.g., limiting knee flexion angle to 0°-135°) are defined within a graph convolutional network (GCN), and backpropagation is used to optimize the feature space distribution.

[0073] The distributed training framework adopts a federated learning architecture. Each edge node calculates the model gradient on the local data set, adds Laplace noise (noise scale ε = 0.1) through the differential privacy mechanism, and performs secure aggregation on the parameter server.

[0074] Furthermore, the step of inputting the temporal feature sequence and the point cloud data into a pre-trained gait recognition model to extract an identity feature vector includes:

[0075] Converting the time series feature sequence into a pulse interval time sequence and accumulating pulse signals;

[0076] Establishing a mapping matrix between the time series feature sequence and the point cloud data according to the pulse signal accumulation, and fusing the time series feature sequence with the point cloud data to obtain a fusion feature;

[0077] The fused features are discretized to generate a feature codebook, an identity identification feature vector of the feature codebook is extracted, and the feature codebook is updated by a clustering algorithm.

[0078] The pulse interval time series conversion adopts a LIF-based neuron model to encode the difference between video frames into pulse firing frequency, with a time resolution of 1ms.

[0079] The mapping matrix is ​​constructed through a cross-modal attention mechanism. Specifically, the spatial correspondence between the visible light feature map (H×W×C) and the radar point cloud (N×3) is calculated, and feature alignment is performed using a learnable weight matrix.

[0080] The generation of fusion features adopts a combination of channel cascading and spatial attention weighting, in which the weight of radar features in occluded areas (such as when the target is blocked by obstacles for more than 50%) is increased to 0.7.

[0081] Discretization is achieved using a vector quantization encoder. The codebook size is set to 512 codewords, each with a dimension of 128. The codebook vectors are dynamically updated using an exponential moving average algorithm. Identity feature vectors are extracted using a nearest neighbor search strategy, calculating the similarity between the query feature and the codewords in Hamming space. The top-3 candidate identities and their confidence scores are returned.

[0082] Furthermore, the multi-target trajectory association based on the identity identification feature vector matching includes calculating the similarity of target features between adjacent frames in the identity identification feature vector, and determining that the targets are the same when the similarity is greater than a preset threshold.

[0083] In the multi-target trajectory association method, the similarity calculation adopts the improved cosine similarity measurement algorithm. Specifically, the feature vector is L2 normalized and then the inner product is calculated. The similarity threshold is set to 0.85 (the ROC curve analysis determines that this value is FRR = 0.3% when FAR = 0.1%).

[0084] Trajectory association utilizes a dual-threshold mechanism: a rapid confirmation is triggered when the similarity of five consecutive frames exceeds 0.9. A secondary verification is performed when the similarity is between 0.85 and 0.9, combining radar Doppler velocity consistency. For brief occlusion scenarios (≤5 seconds), the system initiates trajectory prediction mode, fusing visual observation data with the original radar trajectory using an extended Kalman filter (EKF). Trajectory correction is triggered when the Euclidean distance between the predicted trajectory and the actual detection position exceeds 1.5 meters.

[0085] Furthermore, the output of individual positioning coordinates and behavior classification results in combination with behavior semantics includes:

[0086] Extracting the behavioral semantics of the same target and its displacement trajectory within the target area;

[0087] Obtaining the positioning coordinates of the person according to the displacement trajectory;

[0088] The behavior semantics are analyzed and combined with the positioning coordinates to define the behavior classification results of the person, the behavior classification results including a first classification and a second classification, the first classification including walking, running, and leaving items behind, and the second classification determining wandering behavior based on a stay time threshold.

[0089] Exemplarily, the displacement trajectory is generated by multi-source data fusion: the visual tracking trajectory and the radar point cloud trajectory are weightedly fused using a covariance crossover algorithm (visual weight 0.4, radar weight 0.6), and the trajectory sampling interval is 100ms.

[0090] The mapping of positioning coordinates adopts the perspective projection model to convert the image coordinate system (u, v) into the geographic coordinate system (longitude λ, latitude φ), and the conversion error is controlled within ±0.3m (RMS).

[0091] The first level of behavior classification outputs three basic labels: walking (speed 0.5-1.5m / s), running (>1.5m / s), and item leaving (stationary object duration >60 seconds).

[0092] The second level of classification analyzes the target's residence time in the target area. When the target stays in the target area for more than 120 seconds and the moving distance is less than 1m, it is judged as loitering behavior and an alarm can be triggered (corresponding to different sound and light warning levels).

[0093] All classification results are bound to the positioning coordinates to generate a structured log, which contains four fields: timestamp, WGS84 coordinates, behavior label, and confidence.

[0094] like Figure 2 The figure shows a sensorless positioning system for personnel based on gait recognition, including:

[0095] The data acquisition unit 210 collects two-dimensional video stream data of the target area through a visible light imaging device and collects point cloud data of the target area through a millimeter wave radar;

[0096] The feature generation unit 220 performs three-dimensional reconstruction on the two-dimensional video stream data to generate a temporal feature sequence including spatial coordinates;

[0097] The feature fusion unit 230 inputs the temporal feature sequence and the point cloud data into a pre-trained gait recognition model to extract an identity feature vector;

[0098] The trajectory behavior association unit 240 implements multi-target trajectory association based on the identity identification feature vector matching, and outputs the positioning coordinates of the personnel and the behavior classification results of the personnel in combination with the behavior semantics.

[0099] The electronic device 300 may be a desktop computer, a notebook, a PDA, a cloud server, or other electronic device. The electronic device 300 may include but is not limited to a processor 301 and a memory 302. Those skilled in the art will appreciate that Figure 3 This is merely an example of the electronic device 300 and does not constitute a limitation of the electronic device 300. The electronic device 300 may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the electronic device may also include input and output devices, network access devices, buses, etc.

[0100] The processor 301 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0101] The memory 302 can be an internal storage unit of the electronic device 300, such as a hard disk or memory of the electronic device 300. The memory 302 can also be an external storage device of the electronic device 300, such as a plug-in hard disk equipped on the electronic device 300, a smart memory card (SMC), a secure digital (SD) card, a flash card, etc. Furthermore, the memory 302 can also include both an internal storage unit of the electronic device 300 and an external storage device. The memory 302 is used to store the computer program 303 and other programs and data required by the electronic device. The memory 302 can also be used to temporarily store data that has been output or is about to be output.

[0102] In the embodiments provided in the present disclosure, it should be understood that the disclosed devices / electronic devices and methods can be implemented in other ways. For example, the device / electronic device embodiments described above are merely schematic. For example, the division of modules or units is merely a logical function division. In actual implementation, there may be other division methods. Multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection of devices or units, which may be electrical, mechanical or other forms.

[0103] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present disclosure implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by the processor, it can implement the steps of the above-mentioned various method embodiments. The computer program may include computer program code, which may be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0104] The above embodiments are only used to illustrate the technical solutions of the present disclosure, rather than to limit them. Although the present disclosure has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. 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 various embodiments of the present disclosure, and should all be included in the scope of protection of the present disclosure.

Claims

1. A method for sensorless positioning of personnel based on gait recognition, characterized in that: The following steps are involved: The two-dimensional video stream data of the target area is collected by a visible light imaging device, and the point cloud data of the target area is collected by a millimeter wave radar; Performing three-dimensional reconstruction on the two-dimensional video stream data to generate a temporal feature sequence including spatial coordinates; Inputting the temporal feature sequence and the point cloud data into a pre-trained gait recognition model to extract an identity feature vector; Multi-target trajectory association is achieved based on the identity identification feature vector matching, and the positioning coordinates of the personnel and the behavior classification results of the personnel are output in combination with the behavior semantics.

2. The method according to claim 1, characterized in that The step of performing three-dimensional reconstruction on the two-dimensional video stream data to generate a temporal feature sequence including spatial coordinates includes: Extracting temporally continuous video frames from the two-dimensional video stream data; Performing three-dimensional dynamic reconstruction based on the video frames to generate model data with spatial coordinates; Data enhancement processing is performed on the model data to generate a temporal feature sequence under different environmental interferences, where the different environments include rain, snow, and smoke.

3. The method according to claim 2, characterized in that The method for constructing the pre-training model includes: Construct the first feature extractor including spatiotemporal convolutional network and Transformer structure; Adding the constraint relationship of human skeleton nodes to the first feature extractor to obtain a second feature extractor; Distributed model training is performed on the second feature extractor to obtain a pre-trained model, and the nodes of the second feature extractor update the model parameters based on the differential privacy mechanism.

4. The method according to claim 3, characterized in that Inputting the temporal feature sequence and the point cloud data into a pre-trained gait recognition model to extract an identity feature vector includes: Converting the time series feature sequence into a pulse interval time sequence and accumulating pulse signals; Establishing a mapping matrix between the time series feature sequence and the point cloud data according to the pulse signal accumulation, and fusing the time series feature sequence with the point cloud data to obtain a fusion feature; The fusion features are discretized to generate a feature codebook, an identity identification feature vector of the feature codebook is extracted, and the feature codebook is updated by a clustering algorithm.

5. The method according to claim 1, wherein The multi-target trajectory association based on the identity identification feature vector matching includes calculating the similarity of target features between adjacent frames in the identity identification feature vector, and determining that the targets are the same when the similarity is greater than a preset threshold.

6. The method according to claim 5, wherein the outputting of individual location coordinates and behavior classification results in combination with behavior semantics comprises: Extracting the behavioral semantics of the same target and its displacement trajectory within the target area; Obtaining the positioning coordinates of the person according to the displacement trajectory; The behavior semantics are analyzed and combined with the positioning coordinates to define the behavior classification results of the person, the behavior classification results including a first classification and a second classification, the first classification including walking, running, and leaving items behind, and the second classification determining wandering behavior based on a stay time threshold.

7. A sensorless positioning system for personnel based on gait recognition, characterized in that: include: A data acquisition unit collects two-dimensional video stream data of the target area through a visible light imaging device and collects point cloud data of the target area through a millimeter wave radar; A feature generation unit, performing three-dimensional reconstruction on the two-dimensional video stream data to generate a temporal feature sequence including spatial coordinates; A feature fusion unit, inputting the temporal feature sequence and the point cloud data into a pre-trained gait recognition model to extract an identity feature vector; The trajectory behavior association unit realizes multi-target trajectory association based on the identity identification feature vector matching, and outputs the positioning coordinates of the personnel and the behavior classification results of the personnel in combination with the behavior semantics.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory, characterized in that: When the processor executes the program, a person's sensorless positioning method based on gait recognition is implemented as described in any one of claims 1-6.

9. A computer-readable storage medium storing computer program instructions, characterized in that: When the instructions are executed by the processor, a person's sensorless positioning method based on gait recognition as described in any one of claims 1 to 6 is implemented.

Images