A method for pedestrian and vehicle identification and gate stability control based on millimeter-wave radar

Abstract

This invention relates to a method for pedestrian and vehicle identification and gate stability control based on millimeter-wave radar, comprising the following steps: spatial clustering of point clouds and assignment of target IDs; extraction of three types of features—geometric structure, energy propagation, and principal axis morphology—to construct hierarchical feature vectors; inputting the normalized vectors into a nonlinear probability mapping model to calculate vehicle probability scores; dividing the region according to target distance and constructing a dynamic threshold model to obtain single-frame decision results; then outputting stable decision results through regional weighted multi-frame sliding window statistics; simultaneously establishing a target ID lifecycle management mechanism; and finally outputting gate triggering or anti-collision control signals based on the decision results. This invention, based on millimeter-wave radar point cloud data, achieves reliable pedestrian and vehicle identification through multi-dimensional feature modeling, probability scoring discrimination, and a multi-frame stable decision mechanism, improving the accuracy and stability of pedestrian and vehicle identification and effectively reducing the risk of gate false triggering and accidental collisions.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation and automatic control technology, and in particular to a method for pedestrian and vehicle identification and gate stability control based on millimeter-wave radar. Background Technology

[0002] In parking lot entrances and exits and various passage control scenarios, it is necessary to accurately identify targets within the detection area in order to enable vehicle passage and pedestrian protection against falling objects. With the development of millimeter-wave radar technology, using millimeter-wave radar for target detection has gradually become a common solution. Millimeter-wave radar can acquire information such as the distance, angle and echo signal strength of the target, and describe the distribution of targets within the detection area in the form of point clouds.

[0003] Existing barrier gate detection methods based on millimeter-wave radar typically cluster radar point clouds and make distinctions based on single features such as target distance, number of point clouds, or echo intensity to determine whether the target is a vehicle and trigger barrier gate control. While these methods are simple to implement, they suffer from the following technical problems in practical applications: Because millimeter-wave radar point clouds fluctuate at different distances and under different environmental conditions, relying on a single feature for human and vehicle identification is prone to misjudgment or omission, resulting in insufficient identification accuracy. When making judgments based on single-frame detection results, changes in point cloud distribution can easily cause the judgment results to fluctuate, thus affecting the stability of the barrier gate trigger control. Existing methods lack the ability to adapt to different detection regions, making it difficult to adapt to complex and ever-changing real-world application scenarios. Summary of the Invention

[0004] In view of this, the present invention provides a method for vehicle and pedestrian identification and stable control of a barrier gate based on millimeter-wave radar, which effectively solves the problems of low accuracy and poor stability in the identification of vehicles and pedestrians in the prior art, realizes stable identification of vehicles and pedestrians in the detection area, and outputs barrier gate triggering or anti-collision control signals according to the identification results.

[0005] To achieve the above objectives, the present invention provides a method for pedestrian and vehicle identification and barrier gate stability control based on millimeter-wave radar, comprising the following steps: S1. Collect point cloud data from each frame of millimeter-wave radar and perform spatial clustering processing to extract independent target point cloud sets and assign a unique target ID to each clustered target. S2. Based on the clustered target point cloud set, the target point cloud is divided into three categories according to physical attributes: geometric structure features, energy propagation features, and principal axis morphology features, and a hierarchical feature vector is constructed. S3. Normalize the hierarchical feature vectors and establish a nonlinear probability mapping model constrained by physical priority to calculate the probability score of the target vehicle. S4. Divide the detection space into regions based on the distance to the nearest point of the target and the point cloud density, and construct a region-coupled dynamic threshold model. Compare the vehicle probability score with the corresponding region threshold to obtain the single-frame decision result. S5. Establish a regional weighted multi-frame sliding decision window for the same target ID, perform weighted statistics on the decision results of consecutive frames, output stable vehicle and pedestrian judgment results, and output gate trigger or anti-collision control signals based on the stable judgment results. S501. Establish a region-weighted multi-frame sliding decision window of length N. The expression is: in, This represents the single-frame decision result for the target in the k-th frame; S502. Perform a weighted count on the decision results of consecutive frames, as expressed by: in, Indicates the decision results for consecutive frames. Indicates the first The weights corresponding to each spatial region; S503, when the decision result of consecutive frames Not less than the preset stable decision threshold When the vehicle is stable, output the vehicle determination result and output the vehicle trigger signal to control the gate to open; when the continuous determination results... Less than the preset stable decision threshold When the target is determined to be a pedestrian or a non-vehicle target, an anti-collision protection signal is output to keep the barrier gate closed or to trigger the anti-collision mechanism.

[0006] Preferably, after spatial clustering processing, the first... The first frame A cluster of target point clouds The expression is: in, Indicates the first A point cloud, Represents the spatial coordinates of a point cloud. This represents the echo signal-to-noise ratio of a point cloud. Represents the target point cloud set The number of midpoint clouds; The expression for the center coordinates of the target point cloud set is: The expression for the target center distance is: Preferably, the geometric features include the three-dimensional bounding box size, target volume, and point cloud density; The dimension expression of the three-dimensional bounding box is: in, Indicates the target is Spatial span of dimensions Indicates the target is Spatial span of dimensions Indicates the target is Spatial span of dimensions Indicates the first Frame number The first goal A point cloud coordinate, Indicates the first Frame number The first goal A point cloud coordinate, Indicates the first Frame number The first goal A point cloud coordinate, This indicates taking the maximum function. This represents the function that takes the minimum value. The target volume expression is: in, Indicates the target volume; The point cloud density expression is: in, Represents point cloud density. Represents the target point cloud set The number of midpoint clouds; The geometric feature vector is obtained by normalizing the 3D bounding box size, target volume, and point cloud density, and the expression is: in, express Orientation-normalized size features reflect the target's relative size in the horizontal direction. express Orientation-normalized size features reflect the relative size of the target in the horizontal and vertical directions. express Orientation-normalized size features reflect the relative size of the target height. Represents the normalized point cloud density characteristics. , , They represent direction, direction, Maximum dimension reference in direction. Indicates a reference for point cloud density size; The energy propagation characteristics include average signal-to-noise ratio and distance-compensated signal-to-noise ratio, and the expression for the average signal-to-noise ratio is: in, This represents the average signal-to-noise ratio. Indicates the first Frame number The first goal Signal-to-noise ratio of each point cloud detection point; The distance-compensated signal-to-noise ratio expression is: in, This indicates the distance-compensated signal-to-noise ratio. Indicates the distance compensation factor. Indicates the radial distance from the target to the radar. Indicates the distance decay index; The energy eigenvector is obtained through normalization, and its expression is: in, Represents the energy eigenvector. This represents the maximum statistical value of the signal-to-noise ratio of all point clouds in the current frame, used as the upper bound for energy feature normalization. This represents the minimum statistical value of the signal-to-noise ratio of all point clouds in the current frame, used as the lower bound for energy feature normalization; The principal axis morphology features are determined by calculating the point cloud covariance matrix and its eigenvalues ​​using principal component analysis to determine the principal axis dimensions. The expression for the point cloud covariance matrix is ​​as follows: in, Indicates the first Frame number The first goal A three-dimensional coordinate vector of a point cloud. Indicates the first Frame number The first goal The centroid coordinate vector of a point cloud. Indicates matrix transpose; The eigenvalues ​​are calculated using the following expression: in, Represents the covariance matrix The largest eigenvalue corresponds to the variance along the principal axis of the target point cloud. Represents the covariance matrix The second largest eigenvalue corresponds to the variance of the target point cloud along the second axis. All represent the covariance matrix The minimum eigenvalue corresponds to the variance of the minimum axis direction of the target point cloud; spindle dimensions The expression is: The principal axis eigenvectors are obtained through normalization. The expression is: in, This represents the maximum reference value for the spindle size, used for spindle feature normalization.

[0007] Preferably, the vehicle probability score is obtained through a nonlinear probability mapping model, expressed as: in, This represents a combination of nonlinear probability mapping models for eigenvectors. This represents the Sigmoid mapping bias term, used to adjust the decision boundary for determining vehicle probabilities. This represents the vehicle probability score. , , , , , All represent weight coefficients, and satisfy: =1 and .

[0008] Preferably, in step S4, the detection area is divided by the nearest point distance of the target, wherein the nearest point distance of the target... The expression is: in, Indicates the first Frame number The first goal A point cloud coordinate, Indicates the first Frame number The first goal A point cloud coordinate, Indicates the first Frame number The first goal A point cloud coordinate; Each region corresponds to a different judgment threshold, based on the distance to the nearest point of the target. Construct a region-coupled dynamic threshold model, with the following expression: in, This indicates a dynamic decision threshold for region coupling, which adaptively adjusts according to the point cloud density of the target region. This represents the basic decision threshold, used as the minimum threshold for distinguishing between people and vehicles. This represents the density compensation term, which dynamically adjusts the base threshold based on the target point cloud density. This represents the density compensation coefficient, which controls the degree to which point cloud density affects the threshold. Indicates point cloud density; The single-frame decision result is obtained by comparing the vehicle probability score with the corresponding region threshold. The single-frame decision result is as follows: in, This indicates the decision result for a single frame.

[0009] Preferably, the control method further includes: Construct a target ID lifecycle management model that automatically clears historical sliding window data and resets the judgment status when a target enters a lost state. If the target is not detected within a preset number of consecutive frames, it enters a lost state, that is: in, This represents the number of consecutive frames lost by the target. This indicates the preset target frame loss threshold; The expression for automatically clearing the historical sliding window data, historical judgment results, and judgment status corresponding to the target ID is: in, This represents the sliding window data set corresponding to the target ID, storing the single-frame decision results of the target in the most recent K frames; when the target ID disappears, a zeroing operation is performed to clear all historical decision records to avoid historical data pollution.

[0010] Compared with the prior art, the beneficial effects of the present invention are: This invention constructs a multi-dimensional feature fusion model that includes geometric structural features, energy propagation features, and main axis morphology features to comprehensively analyze millimeter-wave radar point cloud information. Compared with existing technologies that rely on a single feature (such as target distance, point cloud quantity, or echo intensity) for discrimination, this invention can more comprehensively and accurately describe the physical characteristics of the target, thereby significantly improving the recognition accuracy of vehicles and pedestrians and reducing misjudgments and missed judgments. This invention proposes a region-adaptive threshold discrimination mechanism, which divides the detection space according to the distance and position of the target in the detection area and dynamically adjusts the judgment threshold of each region. This mechanism enables human-vehicle discrimination to adapt to changes in point cloud features under different distance conditions (such as differences in point cloud density and echo intensity between near and far distances), thereby improving the system's recognition reliability under different detection areas and environmental conditions. This invention establishes a stable decision mechanism based on target ID through a multi-frame sliding window. It performs weighted statistical analysis on the discrimination results of multiple consecutive frames, effectively overcoming the jitter problem of the judgment result caused by the fluctuation of point cloud distribution when making judgments based on single-frame detection. It significantly reduces the impact of single-frame detection anomalies on the final output and achieves stable and reliable human and vehicle recognition results. This invention deeply integrates stable vehicle and pedestrian recognition results with the gate control strategy: when a vehicle is identified, passage is triggered; when a pedestrian is identified, anti-smashing protection is activated. This effectively reduces the risk of false triggering (pedestrians are mistaken for vehicles, causing the gate to open incorrectly) and false smashing (vehicles are mistaken for pedestrians, causing the gate to smash the vehicle), thus comprehensively improving the safety and reliability of the gate system. Attached Figure Description

[0011] Figure 1 This is the overall flowchart of the pedestrian and vehicle identification and barrier gate control method of the present invention; Figure 2 This is a detailed flowchart of the human-vehicle identification method of the present invention; Figure 3 This is a comparison chart of single-frame probability scores for people and vehicles in this invention. Detailed Implementation

[0012] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0013] This embodiment provides a method for pedestrian and vehicle identification and barrier gate stability control based on millimeter-wave radar, including the following steps: S1. The millimeter-wave radar periodically scans the detection area, outputs point cloud data for each frame, and performs spatial clustering processing. Point clouds belonging to the same target are aggregated, and independent target point cloud sets are extracted to form independent target point cloud sets. A unique target ID is assigned to each clustered target, thereby realizing continuous frame tracking of the target. After spatial clustering processing, the first The first frame A cluster of target point clouds The expression is: in, Indicates the first A point cloud, Represents the spatial coordinates of a point cloud. This represents the echo signal-to-noise ratio of a point cloud. Represents the target point cloud set The number of midpoint clouds; The expression for the center coordinates of the target point cloud set is: The expression for the target center distance is: S2. Based on the clustered target point cloud set, multi-dimensional feature extraction is performed on the target point cloud set. According to the physical attributes (point cloud spatial distribution and echo signal characteristics), it is divided into three categories: geometric structure features, energy propagation features and main axis morphology features, and a hierarchical feature vector is constructed. Geometric features include 3D bounding box size, target volume, and point cloud density, which are used to describe the spatial size and density distribution of the target point cloud. The expression for the size of a 3D bounding box is: in, Indicates the target is Spatial span of dimensions Indicates the target is Spatial span of dimensions Indicates the target is Spatial span of dimensions Indicates the first Frame number The first goal A point cloud coordinate, Indicates the first Frame number The first goal A point cloud coordinate, Indicates the first Frame number The first goal A point cloud coordinate, This indicates taking the maximum function. This represents the function that takes the minimum value. The target volume expression is: in, Indicates the target volume; The expression for point cloud density is: in, Represents point cloud density. Represents the target point cloud set The number of midpoint clouds; The geometric feature vector is obtained by normalizing the 3D bounding box size, target volume, and point cloud density, and the expression is: in, express Orientation-normalized size features reflect the target's relative size in the horizontal direction. express Orientation-normalized size features reflect the relative size of the target in the horizontal and vertical directions. express Orientation-normalized size features reflect the relative size of the target height. Represents the normalized point cloud density characteristics. , , They represent direction, direction, Maximum dimension reference in direction. Indicates a reference for point cloud density size; Energy propagation characteristics include average signal-to-noise ratio (SNR) and range-compensated SNR, which reflect the target echo signal strength and range compensation information. The expression for average SNR is: in, This represents the average signal-to-noise ratio. Indicates the first Frame number The first goal Signal-to-noise ratio of each point cloud detection point; The expression for distance-compensated signal-to-noise ratio is: in, This indicates the distance-compensated signal-to-noise ratio. Indicates the distance compensation factor. Indicates the radial distance from the target to the radar. Indicates the distance decay index; The energy eigenvector is obtained through normalization, and its expression is: in, Represents the energy eigenvector. This represents the maximum statistical value of the signal-to-noise ratio of all point clouds in the current frame, used as the upper bound for energy feature normalization. This represents the minimum statistical value of the signal-to-noise ratio of all point clouds in the current frame, used as the lower bound for energy feature normalization; The principal axis morphology is determined by calculating the point cloud covariance matrix and its eigenvalues ​​using principal component analysis. The expression for the point cloud covariance matrix is ​​as follows: in, Indicates the first Frame number The first goal A three-dimensional coordinate vector of a point cloud. Indicates the first Frame number The first goal The centroid coordinate vector of a point cloud. Indicates matrix transpose; The eigenvalues ​​are calculated using the following expression: in, Represents the covariance matrix The largest eigenvalue corresponds to the variance along the principal axis of the target point cloud. Represents the covariance matrix The second largest eigenvalue corresponds to the variance of the target point cloud along the second axis. All represent the covariance matrix The minimum eigenvalue corresponds to the variance of the minimum axis direction of the target point cloud; spindle dimensions The expression is: The principal axis eigenvectors are obtained through normalization. The expression is: in, This represents the maximum reference value for the spindle dimension, used for spindle feature normalization. S3. Normalize the hierarchical feature vectors and establish a nonlinear probability mapping model constrained by physical priority. Calculate the target vehicle probability score, which reflects the likelihood that the target belongs to the vehicle category. The expression is: in, This represents a combination of nonlinear probability mapping models for eigenvectors. This represents the Sigmoid mapping bias term, used to adjust the decision boundary for determining vehicle probabilities. This represents the vehicle probability score. , , , , , All represent weight coefficients, and satisfy: =1 and ; S4. Based on the nearest point distance to the target and the point cloud density, the detection space is divided into regions, and a region-coupled dynamic threshold model is constructed. Different detection regions correspond to different decision thresholds. The vehicle probability score is compared with the corresponding region threshold to obtain the single-frame decision result. Figure 3 As shown; Target nearest point distance The expression is: in, Indicates the first Frame number The first goal A point cloud coordinate, Indicates the first Frame number The first goal A point cloud coordinate, Indicates the first Frame number The first goal A point cloud coordinate; Each region corresponds to a different judgment threshold, based on the distance to the nearest target point. Construct a region-coupled dynamic threshold model, with the following expression: in, This indicates a dynamic decision threshold for region coupling, which adaptively adjusts according to the point cloud density of the target region. This represents the basic decision threshold, used as the minimum threshold for distinguishing between people and vehicles. This represents the density compensation term, which dynamically adjusts the base threshold based on the target point cloud density. This represents the density compensation coefficient, which controls the degree to which point cloud density affects the threshold. Indicates point cloud density; The single-frame decision result is obtained by comparing the vehicle probability score with the corresponding region threshold. The single-frame decision result is as follows: in, This indicates the decision result for a single frame; S5. In order to improve the stability of the judgment results, the system establishes a multi-frame sliding window judgment mechanism for the same target ID, establishes a regional weighted multi-frame sliding judgment window for the same target ID, performs weighted statistics on the judgment results of consecutive frames, outputs stable people and vehicles judgment results, and outputs gate trigger or anti-collision control signals based on the stable judgment results. S501. Establish a region-weighted multi-frame sliding decision window of length N. The expression is: in, This represents the single-frame decision result for the target in the k-th frame; S502. Perform a weighted count on the decision results of consecutive frames, as expressed by: in, Indicates the decision results for consecutive frames. Indicates the first The weights corresponding to each spatial region; S503, when the decision result of consecutive frames Not less than the preset stable decision threshold When the vehicle is stable, output the vehicle determination result and output the vehicle trigger signal to control the gate to open; when the continuous determination results... Less than the preset stable decision threshold When the target is determined to be a pedestrian or a non-vehicle target, an anti-collision protection signal is output to keep the barrier gate closed or to trigger the anti-collision mechanism. The control signal is output based on the stability decision result, that is: in, Indicates the output control signal. This indicates a stable judgment outcome.

[0014] This embodiment also constructs a target ID lifecycle management model, which automatically clears historical sliding window data and resets the judgment state when the target enters a lost state. If the target is not detected within a preset number of consecutive frames, it enters a lost state, that is: in, This represents the number of consecutive frames lost by the target. This indicates the preset target frame loss threshold; The expression for automatically clearing the historical sliding window data, historical judgment results, and judgment status corresponding to the target ID is: in, This represents the sliding window data set corresponding to the target ID, storing the single-frame decision results of the target in the most recent K frames; when the target ID disappears, a zeroing operation is performed to clear all historical decision records to avoid historical data pollution.

[0015] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for pedestrian and vehicle identification and barrier gate stability control based on millimeter-wave radar, characterized in that, Includes the following steps: S1. Collect point cloud data from each frame of millimeter-wave radar and perform spatial clustering processing to extract independent target point cloud sets and assign a unique target ID to each clustered target. S2. Based on the clustered target point cloud set, the target point cloud is divided into three categories according to physical attributes: geometric structure features, energy propagation features, and principal axis morphology features, and a hierarchical feature vector is constructed. S3. Normalize the hierarchical feature vectors and establish a nonlinear probability mapping model constrained by physical priority to calculate the probability score of the target vehicle. S4. Divide the detection space into regions based on the distance to the nearest point of the target and the point cloud density, and construct a region-coupled dynamic threshold model. Compare the vehicle probability score with the corresponding region threshold to obtain the single-frame decision result. S5. Establish a regional weighted multi-frame sliding decision window for the same target ID, perform weighted statistics on the decision results of consecutive frames, output stable vehicle and pedestrian judgment results, and output gate trigger or anti-collision control signals based on the stable judgment results. S501. Establish a region-weighted multi-frame sliding decision window of length N. The expression is: in, This represents the single-frame decision result for the target in the k-th frame; S502. Perform a weighted count on the decision results of consecutive frames, as expressed by: in, Indicates the decision results for consecutive frames. Indicates the first The weights corresponding to each spatial region; S503, when the decision result of consecutive frames Not less than the preset stable decision threshold When the vehicle is stable, output the vehicle determination result and output the vehicle trigger signal to control the gate to open; when the continuous determination results... Less than the preset stable decision threshold When the target is determined to be a pedestrian or a non-vehicle target, an anti-collision protection signal is output to keep the barrier gate closed or to trigger the anti-collision mechanism.

2. The method for pedestrian and vehicle identification and barrier gate stability control based on millimeter-wave radar according to claim 1, characterized in that, After spatial clustering processing, the first The first frame A cluster of target point clouds The expression is: in, Indicates the first A point cloud, Represents the spatial coordinates of a point cloud. This represents the echo signal-to-noise ratio of a point cloud. Represents the target point cloud set The number of midpoint clouds; The expression for the center coordinates of the target point cloud set is: The expression for the target center distance is: 。 3. The method for pedestrian and vehicle identification and barrier gate stability control based on millimeter-wave radar according to claim 1, characterized in that, The geometric features include the three-dimensional bounding box size, target volume, and point cloud density; The dimension expression of the three-dimensional bounding box is: in, Indicates the target is Spatial span of dimensions Indicates the target is Spatial span of dimensions Indicates the target is Spatial span of dimensions Indicates the first Frame number The first goal A point cloud coordinate, Indicates the first Frame number The first goal A point cloud coordinate, Indicates the first Frame number The first goal A point cloud coordinate, This indicates taking the maximum function. This represents the function that takes the minimum value. The target volume expression is: in, Indicates the target volume; The point cloud density expression is: in, Represents point cloud density. Represents the target point cloud set The number of midpoint clouds; The geometric feature vector is obtained by normalizing the 3D bounding box size, target volume, and point cloud density, and the expression is: in, express Orientation-normalized size features reflect the target's relative size in the horizontal direction. express Orientation-normalized size features reflect the relative size of the target in the horizontal and vertical directions. express Orientation-normalized size features reflect the relative size of the target height. This represents the normalized point cloud density feature. , , They represent direction, direction, Maximum dimension reference in direction. Indicates a reference for point cloud density size; The energy propagation characteristics include average signal-to-noise ratio and distance-compensated signal-to-noise ratio, and the expression for the average signal-to-noise ratio is: in, This represents the average signal-to-noise ratio. Indicates the first Frame number The first goal Signal-to-noise ratio of each point cloud detection point; The distance-compensated signal-to-noise ratio expression is: in, This indicates the distance-compensated signal-to-noise ratio. Indicates the distance compensation factor. Indicates the radial distance from the target to the radar. Indicates the distance decay index; The energy eigenvector is obtained through normalization, and its expression is: in, Represents the energy eigenvector. This represents the maximum statistical value of the signal-to-noise ratio of all point clouds in the current frame, used as the upper bound for energy feature normalization. This represents the minimum statistical value of the signal-to-noise ratio of all point clouds in the current frame, used as the lower bound for energy feature normalization; The principal axis morphology features are determined by calculating the point cloud covariance matrix and its eigenvalues ​​using principal component analysis to determine the principal axis dimensions. The expression for the point cloud covariance matrix is ​​as follows: in, Indicates the first Frame number The first goal A three-dimensional coordinate vector of a point cloud. Indicates the first Frame number The first goal The centroid coordinate vector of a point cloud. Indicates matrix transpose; The eigenvalues ​​are calculated using the following expression: in, Represents the covariance matrix The largest eigenvalue corresponds to the variance along the principal axis of the target point cloud. Represents the covariance matrix The second largest eigenvalue corresponds to the variance of the target point cloud along the second axis. All represent the covariance matrix The minimum eigenvalue corresponds to the variance of the minimum axis direction of the target point cloud; spindle dimensions The expression is: The principal axis eigenvectors are obtained through normalization. The expression is: in, This represents the maximum reference value for the spindle size, used for spindle feature normalization.

4. The method for pedestrian and vehicle identification and barrier gate stability control based on millimeter-wave radar according to claim 1, characterized in that, The vehicle probability score is obtained through a nonlinear probability mapping model, expressed as follows: in, This represents a combination of nonlinear probability mapping models for eigenvectors. This represents the Sigmoid mapping bias term, used to adjust the decision boundary for determining vehicle probabilities. This represents the vehicle probability score. , , , , , All represent weight coefficients, and satisfy: =1 and .

5. The method for pedestrian and vehicle identification and barrier gate stability control based on millimeter-wave radar according to claim 1, characterized in that, In step S4, the detection area is divided according to the nearest point distance of the target. The expression is: in, Indicates the first Frame number The first goal A point cloud coordinate, Indicates the first Frame number The first goal A point cloud coordinate, Indicates the first Frame number The first goal A point cloud coordinate; Each region corresponds to a different judgment threshold, based on the distance to the nearest point of the target. Construct a region-coupled dynamic threshold model, with the following expression: in, This indicates a dynamic decision threshold for region coupling, which adaptively adjusts according to the point cloud density of the target region. This represents the basic decision threshold, used as the minimum threshold for distinguishing between people and vehicles. This represents the density compensation term, which dynamically adjusts the base threshold based on the target point cloud density. This represents the density compensation coefficient, which controls the degree to which point cloud density affects the threshold. Indicates point cloud density; The single-frame decision result is obtained by comparing the vehicle probability score with the corresponding region threshold. The single-frame decision result is as follows: in, This indicates the decision result for a single frame.

6. The method for pedestrian and vehicle identification and barrier gate stability control based on millimeter-wave radar according to claim 1, characterized in that, Control methods also include: Construct a target ID lifecycle management model that automatically clears historical sliding window data and resets the judgment status when a target enters a lost state. If the target is not detected within a preset number of consecutive frames, it enters a lost state, that is: in, This represents the number of consecutive frames lost by the target. This indicates the preset target frame loss threshold; The expression for automatically clearing the historical sliding window data, historical judgment results, and judgment status corresponding to the target ID is: in, This represents the sliding window data set corresponding to the target ID, storing the single-frame decision results of the target in the most recent K frames; when the target ID disappears, a zeroing operation is performed to clear all historical decision records to avoid historical data pollution.

Images