Vehicle safety management method based on smart parking lot

Abstract

The invention relates to the technical field of smart parking lots, in particular to a vehicle safety management method based on a smart parking lot, which comprises the following steps: S1, multi-source data acquisition: acquiring dynamic trajectory data of a target vehicle, parking lot environment parameters and adjacent parking space state data in real time, and generating a multi-source feature vector set; s2, data space-time alignment: carrying out space-time alignment processing on the multi-source feature vector set, and outputting a standardized vehicle behavior sequence; s3, abnormal behavior recognition: inputting the standardized vehicle behavior sequence into an abnormal behavior recognition model, and generating an abnormal behavior characteristic spectrum and a confidence coefficient parameter; s4, risk level assessment: constructing a risk assessment model based on the confidence coefficient parameter and the environmental parameter, and outputting the risk level of the parking lot in the current scene; and S5, security policy triggering: triggering a corresponding security policy execution instruction according to the risk level. According to the invention, potential risks can be effectively prevented, and the safety of vehicles and personnel is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent parking lots, and particularly to a vehicle safety management method based on an intelligent parking lot. Background Art

[0002] With the acceleration of the urbanization process, as an important part of urban infrastructure, the management and safety issues of parking lots have attracted more and more attention. Especially in large-scale intelligent parking lots, where vehicles are dense and the environment is complex, traditional parking lot safety management methods are difficult to cope with the increasing safety risks. Especially in the high-speed and complex urban parking environment, how to monitor the behavior of vehicles in real time and quickly respond to potential safety hazards has become a huge challenge. Therefore, a parking lot safety management system based on intelligent technology, especially a system that can obtain and analyze vehicle behavior, parking lot environment, and parking space status data in real time, has important research significance and application value.

[0003] Currently, most existing parking lot safety management methods rely on a single monitoring system or a static rule judgment mechanism. These methods usually cannot deeply analyze the real-time dynamic behavior of vehicles and cannot identify and handle abnormal behaviors in a timely manner at the initial stage of risk occurrence. For example, existing systems often can only monitor the occupancy of parking spaces through static cameras, lacking the monitoring and analysis of dynamic information such as vehicle movement trajectories, speed changes, and accelerations. At the same time, environmental factors such as light and road surface friction coefficients are often not considered in real time, resulting in misjudgment of risks in complex and harsh environments, thus affecting the efficiency and accuracy of overall safety management. Therefore, the prior art fails to achieve intelligent and dynamic risk assessment and real-time response. Summary of the Invention

[0004] The present invention provides a vehicle safety management method based on an intelligent parking lot, which can effectively improve the safety, real-time response ability, and management efficiency of the parking lot, reduce potential safety hazards, and ensure the safety of personnel, vehicles, and facilities.

[0005] The vehicle safety management method based on an intelligent parking lot includes the following steps:

[0006] S1, multi-source data collection: Real-time obtain the dynamic trajectory data of the target vehicle, parking lot environment parameters, and adjacent parking space status data, and generate a multi-source feature vector set;

[0007] S2, data spatio-temporal alignment: Perform spatio-temporal alignment processing on the multi-source feature vector set, and output a standardized vehicle behavior sequence;

[0008] S3, abnormal behavior recognition: Input the standardized vehicle behavior sequence into an abnormal behavior recognition model, and generate an abnormal behavior feature map and confidence parameters;

[0009] S4, Risk level assessment: Construct a risk assessment model based on confidence parameters and environmental parameters, and output the risk level in the current scenario of the parking lot;

[0010] S5, Safety policy trigger: Trigger the corresponding safety policy execution instruction according to the risk level.

[0011] Optionally, the multi-source data collection in S1 includes:

[0012] S11, Dynamic trajectory data collection: Calculate the three-dimensional motion trajectory of the target vehicle by TOF ranging between the UWB positioning base station and the vehicle-mounted tag, combined with the millimeter-wave radar point cloud data. Specifically, it includes:

[0013] The UWB base station array deployed at the top of the column sends ranging signals at a period of 10 ms, receives the response signals of the vehicle-mounted tag, and calculates the distance d from the base station to the tag through the time difference Δt;

[0014] Use the trilateration algorithm to solve the three-dimensional coordinates (x, y, z) of the vehicle;

[0015] Fuse the radial velocity measurement value v of the millimeter-wave radar r , and eliminate coordinate jitter through Kalman filtering;

[0016] S12, Environmental parameter acquisition: Use multi-sensor fusion technology to collect the environmental characteristics of the parking lot. Specifically, it includes:

[0017] Calculation of ground friction coefficient: Measure the ground pressure distribution P(x, y) through the piezoelectric sensor array, and combine the reflectivity R(λ) of the road surface material detected by the infrared spectrometer to calculate the ground friction coefficient μ;

[0018] Dynamic compensation of light intensity: Perform dynamic compensation on the light intensity of the parking lot;

[0019] S13, Parking space status monitoring: Process the multi-view video stream based on the vehicle contour recognition algorithm of YOLOv5. Specifically, it includes:

[0020] Parking space occupancy determination: When it satisfies continuously for 5 frames , it is determined that the corresponding parking space is occupied, where I edge (x, y) is the edge image gray value of the pixel point (x, y), and I total (x, y) is the original image gray value of the pixel point (x, y);

[0021] Calculation of safety distance: Calculate the safety distance D between vehicles safe , when D safe < 1.5 m, trigger an abnormal mark;

[0022] S14, Feature Vector Generation: Generate feature vectors in a unified format through a spatio-temporal encoder, specifically including:

[0023] Sliding Window Alignment: Define the sliding window length T = 2s to synchronize the time of multi-source data;

[0024] Feature Normalization: Perform normalization on multi-source data using Min-Max normalization;

[0025] Feature Vector Output: Output the multi-source feature vector set Among them, Δx and Δy are the instantaneous trajectory changes of the vehicle, and μ / |v| is the friction-to-speed ratio. is the light intensity gradient, and N danger is the number of abnormal adjacent vehicles satisfying D safe < 0.5m.

[0026] Optionally, the data spatio-temporal alignment in S2 includes:

[0027] S21, Multi-source Timestamp Calibration: Construct a unified synchronous time axis based on sensor clock deviation compensation, specifically including:

[0028] Attach local timestamps t uwb , t radar , t cam to the UWB positioning data, millimeter-wave radar point cloud data, and camera data respectively;

[0029] Calculate the clock deviation compensation amount Δt using the PTP protocol;

[0030] Generate a unified reference time axis T sync = t sensor + Δt, where t sensor is the original timestamp of the sensor data;

[0031] S22, Spatial Coordinate System Unification: Establish a global coordinate system for the parking lot and complete the mapping of multi-source data, specifically including:

[0032] Define the origin of the global northeast celestial coordinate system as O(0,0,0) and set the origin at the geometric center of the parking lot;

[0033] Optimize the UWB base station coordinates using the Levenberg-Marquardt method;

[0034] Convert the camera data to the global coordinate system through a perspective transformation matrix;

[0035] S23, Motion Trajectory Interpolation and Reconstruction: Use adaptive cubic spline interpolation to perform missing compensation and continuity repair on the trajectory data, and introduce speed limit conditions to optimize the curve shape;

[0036] S24, Environmental Feature Association Modeling: Construct a vehicle behavior-environment coupling matrix, define an environmental influence attenuation function w(r), and calculate the comprehensive environmental parameter E(t);

[0037] S25, Standardized Sequence Generation: Perform spatio-temporal dimension normalization processing to generate a standardized vehicle behavior sequence.

[0038] Optionally, the abnormal behavior recognition in S3 includes:

[0039] S31, Spatio-Temporal Feature Map Generation: Use a spatio-temporal graph convolutional network (ST-GCN) to extract features from the vehicle behavior sequence. By defining key motion nodes (such as headlights, wheels, etc.), combined with multi-scale spatio-temporal convolutional kernels, capture the three-dimensional trajectory of the vehicle and time changes, and finally generate a behavior feature map that fuses spatio-temporal characteristics;

[0040] S32, Dynamic Threshold Abnormality Judgment: Introduce an adaptive dynamic threshold mechanism, adjust the sensitivity of the anomaly detection threshold according to environmental parameters (such as light, friction coefficient), extract anomaly indicators through max pooling, and calculate the confidence parameter.

[0041] Optionally, the spatio-temporal feature map generation in S31 includes:

[0042] S311, Definition of Key Motion Nodes: Define a vehicle skeleton model to represent the key motion nodes of vehicle movement, including chassis motion nodes, body feature nodes, and environmental interaction nodes. The chassis motion nodes include the left front wheel center N1, the right front wheel center N2, the left rear wheel center N3, and the right rear wheel center N4. The body feature nodes include the roof center N5, the midpoints of the front and rear bumpers N6, N7, the hinge points of the four car doors N8, N9, N 10 、N 11 , and the environmental interaction nodes include the centroid of the front headlight group N 12 、the centroid of the taillight group N 13 、the license plate center N 14 、the charging port position N 15 , and the key motion nodes include three-dimensional coordinates (x i , y i , z i ), velocity vector acceleration

[0043] S312, Spatial Graph Convolution Modeling: Construct a spatial association matrix between vehicle components and perform graph convolution. Define the connection relationship between vehicle components through the adjacency matrix, considering physical connection and motion similarity. The graph convolution layer updates the node features to capture the spatial association and motion laws between vehicle components, thereby extracting features of spatio-temporal correlation;

[0044] S313, Multi-scale temporal convolution: By using dilated convolution, capture behavioral features at different time scales;

[0045] S314, Spatiotemporal feature fusion: Through a cross-modal attention mechanism, fuse spatial features and temporal features to form the final spatiotemporal feature map.

[0046] Optionally, the dynamic threshold anomaly determination in S32 includes:

[0047] S321, Feature channel pooling: Perform multi-dimensional max pooling operation on the spatiotemporal feature map to extract key anomaly indicators, including the acceleration channel c acc and the yaw angle channel c yaw ;

[0048] S322, Dynamic threshold calculation: Construct a dynamically determined anomaly detection threshold that adapts to environmental parameters;

[0049] S323, Confidence parameter generation: Calculate the confidence level C based on the relative deviation between the key anomaly indicator and the anomaly detection threshold.

[0050] Optionally, the risk level assessment in S4 includes:

[0051] S41, Risk factor fusion modeling: Construct a coupled evaluation model of the confidence parameter and the environmental parameter, and generate a dynamic risk factor R core ;

[0052] S42, Traffic flow density compensation: Introduce the real-time traffic flow density ρ for dynamic adjustment of the dynamic risk factor;

[0053] S43, Time decay correction: Apply a time decay constraint to sudden risk events to generate the final risk value R final ;

[0054] S44, Risk level classification: Based on the generated final risk value R final , classify the risk level, including low risk, medium risk, high risk, and critical risk.

[0055] Optionally, the risk level classification in S44 includes:

[0056] S441, Low risk: When R final ≤50%, it indicates that the current scenario of the parking lot is of low risk;

[0057] S442, Medium risk: When 50% < R final ≤120%, it indicates that the current scenario of the parking lot is of medium risk;

[0058] S443, High risk: When 120% < R finalWhen it is ≤ 250%, it indicates that the current scenario of the parking lot is a medium risk;

[0059] S444, critical risk: When R final > 250%, it indicates that the current scenario of the parking lot is a critical risk.

[0060] Optionally, the triggering of the security policy in S5 includes:

[0061] S51, low-risk policy: When the risk level is low risk, send a voice prompt to the in-vehicle terminal to remind the driver to pay attention, and execute it using the on-board unit (OBU) of the target vehicle;

[0062] S52, medium-risk policy: When the risk level is medium risk, activate the audible and visual warning devices within 3 meters around, and enhance the warning effect through the intelligent roadside unit (RSU);

[0063] S53, high-risk policy: When the risk level is high risk, generate an obstacle avoidance path, and control the intelligent ground lock to rise through the AR navigation system;

[0064] S54, critical-risk policy: When the risk level is high risk, link the charging piles in adjacent parking spaces to cut off the power, and start the physical isolation device, including hydraulic isolation piers for physical safety isolation.

[0065] Advantages of the present invention:

[0066] In the present invention, through the vehicle safety management method based on multi-source data collection and intelligent analysis, the dynamic trajectory data of the target vehicle, the parking lot environment parameters, and the parking space status data can be obtained in real time, so as to accurately identify the abnormal behaviors of the vehicle and conduct risk assessment in a timely manner. By using the space-time alignment and feature fusion technology, the present invention can ensure a comprehensive analysis of vehicle behaviors. Combining the abnormal behavior recognition model with the environment parameters, the risk level of the current scenario of the parking lot can be evaluated in real time.

[0067] In the present invention, through the evaluated risk level, different security policies can be triggered to ensure that the risks in the parking lot are effectively controlled. When a low risk is detected, the system will send a voice prompt to the in-vehicle terminal; when the risk level increases, the audible and visual warning devices will be activated or the intelligent ground lock will be controlled, and even physical isolation measures will be started in critical situations. Through this intelligent policy execution mechanism, potential risks can be effectively prevented and the safety of vehicles and personnel can be ensured. Description of the Drawings

[0068] To more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only those of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0069] Figure 1 Schematic diagram of the management method process for the embodiments of the present invention;

[0070] Figure 2 Schematic diagram of the risk level assessment for the embodiments of the present invention. Detailed implementation manners

[0071] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. At the same time, it should be noted here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments. For some well-known technologies, those skilled in the art can also adopt other alternative methods for implementation; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0072] It should be pointed out that in the specification, when referring to "an embodiment", "embodiments", "exemplary embodiments", "some embodiments", etc., it indicates that the described embodiments may include specific features, structures or characteristics, but not necessarily every embodiment includes such specific features, structures or characteristics. In addition, when combining embodiments to describe specific features, structures or characteristics, implementing such features, structures or characteristics in combination with other embodiments (whether explicitly described or not) should be within the knowledge of those skilled in the relevant art.

[0073] Generally, terms can be understood at least in part from their use in the context. For example, at least in part depending on the context, the term "one or more" used herein can be used to describe any feature, structure or characteristic in a singular sense, or can be used to describe a combination of features, structures or characteristics in a plural sense. In addition, the term "based on" can be understood as not necessarily intended to convey a set of exclusive factors, but instead, at least in part depending on the context, allowing the existence of other factors that may not be explicitly described.

[0074] As Figure 1 - Figure 2 shown, the vehicle safety management method based on the intelligent parking lot includes the following steps:

[0075] S1, Multi-source data collection: Real-time obtain the dynamic trajectory data of the target vehicle, the parking lot environment parameters and the adjacent parking space status data, and generate a multi-source feature vector set;

[0076] S2, Data spatio-temporal alignment: Perform spatio-temporal alignment processing on the multi-source feature vector set, and output a standardized vehicle behavior sequence;

[0077] S3, Abnormal behavior recognition: Input the standardized vehicle behavior sequence into the abnormal behavior recognition model to generate an abnormal behavior feature map and confidence parameters;

[0078] S4, Risk level assessment: Build a risk assessment model based on the confidence parameters and environmental parameters, and output the risk level in the current parking lot scenario;

[0079] S5, Safety policy trigger: Trigger the corresponding safety policy execution instruction according to the risk level.

[0080] The multi-source data collection in S1 includes:

[0081] S11, Dynamic trajectory data collection: Calculate the three-dimensional motion trajectory of the target vehicle by the TOF ranging between the UWB positioning base station and the vehicle-mounted tag, combined with the millimeter-wave radar point cloud data, specifically including:

[0082] The UWB base station array deployed at the top of the column sends ranging signals at a period of 10 ms, receives the response signals of the vehicle-mounted tag, and calculates the distance d from the base station to the tag through the time difference Δt, expressed as:

[0083]

[0084] where c is the speed of light;

[0085] Use the trilateration algorithm to solve the three-dimensional coordinates (x, y, z) of the vehicle, expressed as:

[0086]

[0087] where (x1, y1, z1), (x2, y2, z2), (x3, y3, z3) are the coordinates of the 1st, 2nd, and 3rd base stations respectively, and d1, d2, d3 are the corresponding ranging values;

[0088] Fuse the radial velocity measurement value v of the millimeter-wave radar r , and eliminate the coordinate jitter through Kalman filtering, expressed as:

[0089]

[0090] where is the estimated value of the vehicle position after Kalman filtering, α, β are the coordinate fusion weight coefficients, is the vehicle position coordinate at the current moment solved by UWB, and θ is the angle between the radar beam and the vehicle movement direction;

[0091] S12, Environmental parameter acquisition: Use multi-sensor fusion technology to collect the environmental characteristics of the parking lot, specifically including:

[0092] Calculation of ground friction coefficient: Measure the ground pressure distribution P(x, y) through a piezoelectric sensor array, and combine with an infrared spectrometer to detect the reflectivity R(λ) of the road surface material, and calculate the ground friction coefficient μ, which is expressed as:

[0093]

[0094] where max(P) is the maximum value in the pressure array, P std is the standard deviation of the pressure data, and w1, w2 are the weight coefficients in the calculation of friction coefficient fusion;

[0095] Dynamic compensation of light intensity: Perform dynamic compensation on the light intensity in the parking lot, which is expressed as:

[0096]

[0097] where L real is the actual light intensity after dynamic compensation, L sensor is the original reading of the light sensor, L dark is the dark current noise of the sensor, t exposure is the exposure time of the camera, and K is the exposure compensation calibration coefficient;

[0098] S13, Parking space status monitoring: Process multi-view video streams based on the vehicle contour recognition algorithm of YOLOv5, specifically including:

[0099] Parking space occupancy determination: When 5 consecutive frames satisfy , it is determined that the corresponding parking space is occupied. Among them, I edge (x, y) is the edge image gray value of the pixel point (x, y), and I total (x, y) is the original image gray value of the pixel point (x, y);

[0100] Calculation of safety distance: Calculate the safety distance D safe between vehicles. When D safe < 1.5m, an abnormal mark is triggered. The safety distance D safe is expressed as:

[0101]

[0102] where (x i , y i ) is the central position coordinate of the i-th vehicle, and (x j , y j ) is the central position coordinate of the j-th vehicle;

[0103] S14, Feature vector generation: Generate a feature vector in a unified format through a spatio-temporal encoder, specifically including:

[0104] Sliding window alignment: Define the sliding window length T = 2s, and perform time synchronization on multi-source data, expressed as:

[0105]

[0106] where t is the current time point, Traj(k) is the subset of trajectory data at time k, Env(k) is the subset of environmental parameters at time k, and Slot(k) is the subset of parking space status parameters at time k;

[0107] Feature normalization: Use Min-Max normalization to normalize multi-source data, expressed as:

[0108]

[0109] where is the i-th normalized eigenvalue, F i is the original eigenvalue, F min and F max are the minimum and maximum values of the original eigenvalue respectively;

[0110] Feature vector output: Output the multi-source feature vector set where Δx and Δy are the instantaneous trajectory changes of the vehicle, μ / |v| is the friction-to-speed ratio, is the light intensity gradient, N danger is the number of abnormal neighboring vehicles satisfying D safe < 0.5m.

[0111] The data spatio-temporal alignment in S2 includes:

[0112] S21, multi-source timestamp calibration: Construct a unified synchronization time axis based on sensor clock deviation compensation, specifically including:

[0113] Attach local timestamps t uwb , t radar , and t cam to the UWB positioning data, millimeter-wave radar point cloud data, and camera data respectively;

[0114] Use the PTP protocol to calculate the clock deviation compensation amount Δt, expressed as:

[0115]

[0116] where t1 and t4 are the master-slave clock synchronization message sending and receiving times, t2 and t3 are the slave-master clock response times, and α ′ is the network delay correction factor, and σ delay is the historical delay standard deviation;

[0117] Generate a unified reference time axis Tsync = t sensor + Δt, where t sensor is the original timestamp of the sensor data;

[0118] S22, Spatial coordinate system unification: Establish a global coordinate system for the parking lot and complete the mapping of multi-source data, specifically including:

[0119] Define the origin of the global northeast celestial coordinate system as O(0, 0, 0), and set the origin at the geometric center of the parking lot;

[0120] Perform Levenberg-Marquardt optimization on the UWB base station coordinates, expressed as:

[0121]

[0122] where a i is the initial coordinate of the base station, d i is the measured distance, λ = 1e ；4 is the damping factor, d i is the ranging value, N is the number of base stations in the optimization problem, x is the coordinate point to be optimized, and J is the Jacobian matrix;

[0123] Convert the camera data to the global coordinate system through the perspective transformation matrix, expressed as:

[0124]

[0125] where (u, v) are the pixel coordinates, (X, Y) are the mapped ground coordinates, and H is the homography matrix;

[0126] S23, Motion trajectory interpolation and reconstruction: Use adaptive cubic spline interpolation to perform missing compensation and continuity repair on the trajectory data, and introduce speed limit conditions to optimize the curve shape, expressed as:

[0127] S j (t) = a j (t - t j ) 3 + b j (t - t j ) 2 + c j (t - t j ) + d j ;

[0128]

[0129]

[0130] where S j (t) is the j-th cubic spline interpolation function, aj , b j , c j , d j are the coefficients of the j-th interpolation function, and t j is the start time of the j-th interpolation interval. is the k-th derivative of the j-th interpolation function at t j:1 . is the k-th derivative of the (j + 1)-th interpolation function at t j:1 . is the acceleration of the interpolation trajectory, a max is the maximum allowable acceleration, dθ / dt is the rate of change of the heading angle, and θ is the vehicle heading angle;

[0131] S24, Environmental Feature Association Modeling: Construct a vehicle behavior-environment coupling matrix, define an environmental influence attenuation function w(r), and calculate the comprehensive environmental parameter E(t), expressed as:

[0132]

[0133] where r is the distance from the vehicle to the environmental monitoring point, μ(r) is the friction coefficient measured at this point, σ = 3m is the influence radius, μ0 = 0.6 is the reference friction coefficient, and L k (t) is the light intensity at the k-th point, and μ k (t) is the friction coefficient at the k-th point, and K is the number of monitoring points participating in the calculation;

[0134] S25, Standardized Sequence Generation: Perform spatio-temporal dimension normalization processing to generate a standardized vehicle behavior sequence, specifically including:

[0135] Downsample the time to a unified frequency of 10Hz, using anti-aliasing sinc interpolation, expressed as:

[0136]

[0137] where is the downsampled value at the sampling frequency of , T is the target unified sampling period, Δt ′ is the original sampling period, M is the filter order, and x(mΔt ′ ) is the signal value at the m-th sampling point in the original data;

[0138] Map the coordinate values to the interval [-1, 1] through spatial normalization, expressed as:

[0139]

[0140] where p is the original vehicle three-dimensional coordinate, p min , p maxis the historical statistical extreme value, s range = 2, o bias = -1 are the mapping parameters.

[0141] The abnormal behavior recognition in S3 includes:

[0142] S31, Spatiotemporal Feature Map Generation: Use the Spatiotemporal Graph Convolutional Network (ST-GCN) to extract features from the vehicle behavior sequence. By defining key motion nodes (such as headlights, wheels, etc.), combined with multi-scale spatiotemporal convolutional kernels, capture the three-dimensional trajectory of the vehicle and the time variation, and finally generate a behavior feature map that fuses spatiotemporal characteristics;

[0143] S32, Dynamic Threshold Abnormality Judgment: Introduce an adaptive dynamic threshold mechanism, adjust the sensitivity of the anomaly detection threshold according to environmental parameters (such as light, friction coefficient), extract the anomaly index through max pooling, improve the anomaly detection rate in harsh environments, and reduce false alarms in sunny days with high friction. The anomaly detection rate in rainy days with low illuminance is increased to 89%, and the false alarm rate is reduced by 41%, and the confidence parameter is calculated.

[0144] The spatiotemporal feature map generation in S31 includes:

[0145] S311, Definition of Key Motion Nodes: Define the key motion nodes representing the vehicle motion using the vehicle bone model, including the chassis motion node, body feature node, and environmental interaction node. The chassis motion node includes the left front wheel center N1, right front wheel center N2, left rear wheel center N3, and right rear wheel center N4. The body feature nodes include the roof center N5, the midpoints of the front and rear bumpers N6, N7, the hinge points of the four car doors N8, N9, N 10 、N 11 , and the environmental interaction node includes the centroid of the front headlight group N 12 、the centroid of the taillight group N 13 、the license plate center N 14 、the charging port position N 15 , and the key motion nodes include three-dimensional coordinates (x i , y i , z i ), velocity vector acceleration

[0146] S312, Spatial Graph Convolution Modeling: Construct the spatial correlation matrix between vehicle components and perform graph convolution. Define the connection relationship between vehicle components through the adjacency matrix, considering physical connection and motion similarity. The graph convolution layer updates the node features to capture the spatial correlation and motion law between vehicle components, so as to extract the features of spatiotemporal correlation, expressed as:

[0147]

[0148] Among them, A ij is an adjacency matrix element, representing the correlation between nodes i and j, and σ ′ is the speed similarity determination coefficient, is the velocity vector of the node;

[0149]

[0150] Λ ii = ∑ j (A + I) ij ;

[0151] Among them, H (l:1) is the graph convolution feature matrix of the (l + 1)-th layer, Λ is the diagonalized degree matrix, A is the adjacency matrix, I is the identity matrix, H (l) is the graph convolution feature matrix of the l-th layer, and W (l) is the trainable weight matrix of the l-th layer;

[0152] S313, multi-scale temporal convolution: By using dilated convolution, the behavioral features at different time scales are captured, expressed as:

[0153]

[0154] Among them, T (l:1) (t) is the output feature of the temporal convolution of the (l + 1)-th layer, the eigenvalue at time step t, and W (l) (τ) is the convolutional kernel weight of the l-th layer, and H (l) (t - d l ·τ) is the input feature matrix of the l-th layer, representing the feature at time step t - d l ·τ, and k l is the width of the convolutional kernel of the l-th layer, controlling the convolution span, and k l = 2 l:1 , and d l is the dilation factor of the l-th layer, and d l = 2 l ;

[0155] S314, spatio-temporal feature fusion: By means of a cross-modal attention mechanism, the spatial features and temporal features are fused to form the final spatio-temporal feature map, expressed as:

[0156]

[0157] Among them, M t,i,c is the output after spatio-temporal feature fusion, representing the fused feature at time step t, spatial node i, and feature channel c, and α ij is the attention weight, ReLU is the activation function, is the spatial feature and the temporal feature The splicing of W f is the fusion weight matrix, and LeakyReLU is the Leaky ReLU activation function. is the spatial feature and The splicing of is the spatial feature and The splicing of.

[0158] The dynamic threshold anomaly determination in S32 includes:

[0159] S321, Feature channel pooling: Perform multi-dimensional maximum pooling operations on the spatio-temporal feature map to extract key anomaly indicators, including the acceleration channel c acc and the yaw angle channel c yaw , expressed as:

[0160]

[0161] where F max is the maximum anomaly indicator after the pooling operation, is the acceleration channel eigenvalue at the t-th time step and the i-th node, is the yaw angle channel eigenvalue at the t-th time step and the i-th node, λ = 0.6 is the weight coefficient of the yaw angle, and T is the time window length;

[0162] S322, Dynamic threshold calculation: Construct a dynamic decision anomaly detection threshold that adapts to environmental parameters, expressed as:

[0163]

[0164] where θ dynamic is the anomaly detection threshold, θ base is the basic threshold for sunny standard scenarios, w L is the weight affected by light intensity, L curvent is the light intensity of the current scenario, L max is the maximum calibration value, representing the upper limit of light intensity, w μ is the weight affected by the friction coefficient, μ base is the reference friction coefficient under dry road surfaces, μ current is the friction coefficient measured in real time. When L curvent decreases or μ current decreases, θ dynamic decreases to improve the detection sensitivity. When L curvent increases or μ current increases, θ dynamic increases to reduce the false alarm rate;

[0165] S323, Confidence Parameter Generation: Calculate the confidence level C based on the relative deviation between the key anomaly indicator and the anomaly detection threshold, expressed as:

[0166]

[0167] Among them, C ∈ [0%, 300%] is the classification trigger warning (C > 100% is a high-risk event). When μ current < 0.3, apply a 1.2-fold gain compensation to C.

[0168] The risk level assessment in S4 includes:

[0169] S41, Risk Factor Fusion Modeling: Construct a coupled evaluation model of the confidence parameter and the environmental parameter, and generate the dynamic risk factor R core , expressed as:

[0170]

[0171] Among them, η is the environmental coupling coefficient, k ′ is the attenuation coefficient, R core is the initially calculated risk factor, L norm is the normalized light intensity, is the environmental impact gain, is the non-linear adjustment factor;

[0172] S42, Traffic Flow Density Compensation: Introduce the real-time traffic flow density ρ for dynamic adjustment of the dynamic risk factor, expressed as:

[0173]

[0174] Among them, R adjusted is the dynamically adjusted risk factor, γ is the weight coefficient of the traffic flow density, ρ base is the reference traffic flow density, ρ max is the maximum traffic flow density;

[0175] S43, Time Decay Correction: Apply a time decay constraint to the sudden risk event to generate the final risk value R final , expressed as:

[0176]

[0177] Among them, R final (t) is the final risk value, λ is the time decay rate, Δt is the time difference between the current moment and the start time of the risk event, is the risk value of the i-th historical risk event, δ is the historical risk attenuation factor, s is the number of historical risk events;

[0178] S44, Risk level classification: Based on the generated final risk value R final , classify the risk levels, including low risk, medium risk, high risk, and critical risk.

[0179] The risk level classification in S44 includes:

[0180] S441, Low risk: When R final ≤50%, it indicates that the current scenario of the parking lot is of low risk;

[0181] S442, Medium risk: When 50% < R final ≤120%, it indicates that the current scenario of the parking lot is of medium risk;

[0182] S443, High risk: When 120% < R final ≤250%, it indicates that the current scenario of the parking lot is of medium risk;

[0183] S444, Critical risk: When R final >250%, it indicates that the current scenario of the parking lot is of critical risk.

[0184] The triggering of the safety policies in S5 includes:

[0185] S51, Low risk policy: When the risk level is low risk, send a voice prompt to the in-vehicle terminal to remind the driver to pay attention, and execute it using the on-board unit (OBU) of the target vehicle;

[0186] S52, Medium risk policy: When the risk level is medium risk, activate the acoustic and optical warning devices within 3 meters around, and enhance the warning effect through the intelligent roadside unit (RSU);

[0187] S53, High risk policy: When the risk level is high risk, generate an obstacle avoidance path, and control the intelligent ground lock to rise through the AR navigation system to ensure vehicle safety;

[0188] S54, Critical risk policy: When the risk level is high risk, link the charging piles of adjacent parking spaces to cut off the power, and start the physical isolation device, including hydraulic isolation piers, for physical safety isolation to ensure maximum protection.

[0189] The present invention covers any substitutions, modifications, equivalent methods, and solutions made within the essence and scope of the present invention. To enable the public to have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, and those skilled in the art can fully understand the present invention without the description of these details. Additionally, well-known methods, processes, procedures, components, and circuits are not described in detail to avoid unnecessary confusion to the essence of the present invention.

[0190] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A vehicle safety management method based on an intelligent parking lot, characterized in that, It includes the following steps: S1, Multi-source data collection: Real-time obtain the dynamic trajectory data of the target vehicle, parking lot environment parameters, and adjacent parking space status data, and generate a multi-source feature vector set; S2, Data spatio-temporal alignment: Perform spatio-temporal alignment processing on the multi-source feature vector set and output a standardized vehicle behavior sequence; S3, Abnormal behavior recognition: Input the standardized vehicle behavior sequence into the abnormal behavior recognition model to generate an abnormal behavior feature map and confidence parameters; S4, Risk level assessment: Build a risk assessment model based on the confidence parameters and environment parameters and output the risk level under the current parking lot scenario; S5, Safety policy trigger: Trigger the corresponding safety policy execution instruction according to the risk level.

2. The vehicle safety management method based on an intelligent parking lot according to claim 1, wherein The multi-source data collection in S1 includes: S11, Dynamic trajectory data collection: Calculate the three-dimensional motion trajectory of the target vehicle through the TOF ranging between the UWB positioning base station and the vehicle-mounted tag, combined with the millimeter-wave radar point cloud data. Specifically, it includes: The UWB base station array deployed at the top of the column sends ranging signals at a period of 10 ms, receives the response signals of the vehicle-mounted tag, and calculates the distance d from the base station to the tag through the time difference Δt; Use the trilateration algorithm to solve the three-dimensional coordinates (x, y, z) of the vehicle; Fuse the radial velocity measurement value v of the millimeter-wave radar r , and eliminate coordinate jitter through Kalman filtering; S12, Environment parameter acquisition: Use multi-sensor fusion technology to collect the parking lot environment characteristics. Specifically, it includes: Ground friction coefficient calculation: Measure the ground pressure distribution P(x, y) through the piezoelectric sensor array, and combine the infrared spectrum analyzer to detect the road surface material reflectivity R(λ) to calculate the ground friction coefficient μ; Dynamic compensation of light intensity: Perform dynamic compensation on the light intensity in the parking lot; S13, Parking space status monitoring: Process the multi-view video stream based on the vehicle contour recognition algorithm of YOLOv5. Specifically, it includes: Parking space occupancy determination: When 5 consecutive frames satisfy , it is determined that the corresponding parking space is occupied, where I edge (x,y) is the edge image gray value of the pixel point (x,y), and I total (x,y) is the original image gray value of the pixel point (x,y); Safety distance calculation: Calculate the safety distance D between vehicles safe , when D safe < 1.5 m, trigger an abnormal mark; S14, Feature vector generation: Generate feature vectors in a unified format through a spatio-temporal encoder. Specifically, it includes: Sliding window alignment: Define the sliding window length T = 2 s to synchronize the multi-source data in time; Feature normalization: Perform normalization processing on the multi-source data using the Min-Max normalization method; Feature vector output: Output a multi-source feature vector set where Δx and Δy are the instantaneous trajectory changes of the vehicle, and μ / |v| is the friction-to-speed ratio is the light intensity gradient, and N danger is the number of abnormal adjacent vehicles satisfying D safe < 0.5m 3. The vehicle safety management method based on an intelligent parking lot according to claim 1, characterized in that The data spatio-temporal alignment in S2 includes: S21, Multi-source timestamp calibration: Build a unified synchronization time axis based on sensor clock deviation compensation. Specifically, it includes: Stamp the local timestamps t on the UWB positioning data, millimeter-wave radar point cloud data, and camera data respectively uwb 、t radar 、t cam ; Use the PTP protocol to calculate the clock deviation compensation amount Δt; Generate a unified reference timeline T sync = t sensor + Δt, where t sensor is the original timestamp of the sensor data; S22, Unification of spatial coordinate systems: Establish a global coordinate system for the parking lot and complete the mapping of multi-source data. Specifically, it includes: Define the origin of the global northeast celestial coordinate system as O(0, 0, 0) and set the origin at the geometric center of the parking lot; Perform Levenberg-Marquardt optimization on the UWB base station coordinates; Convert the camera data to the global coordinate system through the perspective transformation matrix; S23, Motion trajectory interpolation and reconstruction: Use adaptive cubic spline interpolation to perform missing compensation and continuity repair on the trajectory data, and introduce speed limit conditions to optimize the curve shape; S24, Environment feature correlation modeling: Build a vehicle behavior-environment coupling matrix, define the environmental influence attenuation function w(r), and calculate the comprehensive environment parameter E(t); S25, Standardized sequence generation: Perform spatio-temporal dimension normalization processing to generate a standardized vehicle behavior sequence.

4. The vehicle safety management method based on an intelligent parking lot according to claim 3, wherein, The abnormal behavior recognition in S3 includes: S31, Spatio-temporal feature map generation: Use a spatio-temporal graph convolutional network to extract features from the vehicle behavior sequence. By defining key motion nodes and combining multi-scale spatio-temporal convolutional kernels, capture the three-dimensional trajectory and time variation of the vehicle, and finally generate a behavior feature map that fuses spatio-temporal characteristics; S32, Dynamic threshold abnormal determination: Introduce an adaptive dynamic threshold mechanism, adjust the sensitivity of the abnormal detection threshold according to environmental parameters, extract abnormal indicators through max pooling, and calculate the confidence parameter.

5. The vehicle safety management method based on an intelligent parking lot according to claim 4, characterized in that The spatio-temporal feature map generation in S31 includes: S311, Definition of Key Motion Nodes: Define the key motion nodes that represent the vehicle motion in the vehicle skeleton model, including chassis motion nodes, body feature nodes, and environment interaction nodes. The chassis motion nodes include the left front wheel center N1, the right front wheel center N2, the left rear wheel center N3, and the right rear wheel center N4. The body feature nodes include the roof center N5, the midpoints N6 and N7 of the front and rear bumpers, and the four door hinge points N8, N9, N 10 、N 11 , and the environment interaction nodes include the centroid N 12 of the front headlight group, the centroid N 13 of the rear taillight group, the license plate center N 14 , and the charging port position N 15 . The key motion nodes include three-dimensional coordinates (x i , y i , z i ), velocity vector acceleration S312, Spatial graph convolution modeling: Construct a spatial association matrix between vehicle components and perform graph convolution. Define the connection relationship between vehicle components through the adjacency matrix, consider physical connection and motion similarity, and the graph convolution layer updates the node features to capture the spatial association and motion law between vehicle components, so as to extract features of spatio-temporal correlation; S313, Multi-scale time convolution: Capture behavior features at different time scales by using dilated convolution; S314, Spatio-temporal feature fusion: Through a cross-modal attention mechanism, fuse spatial features and time features to form the final spatio-temporal feature map.

6. The vehicle safety management method based on an intelligent parking lot according to claim 5, wherein The dynamic threshold abnormal determination in S32 includes: S321, Feature Channel Pooling: Perform multi-dimensional maximum pooling operations on the spatio-temporal feature map to extract key anomaly indicators, including the acceleration channel c acc and the yaw angle channel c yaw ; S322, Dynamic threshold calculation: Construct a dynamic determination abnormal detection threshold that adapts to environmental parameters; S323, Confidence parameter generation: Calculate the confidence C based on the relative deviation between the key abnormal indicator and the abnormal detection threshold.

7. The vehicle safety management method based on an intelligent parking lot according to claim 6, wherein The risk level assessment in S4 includes: S41, Risk factor fusion modeling: Construct a coupled evaluation model of confidence parameters and environmental parameters, and generate a dynamic risk factor R core ; S42, Traffic flow density compensation: Introduce the real-time traffic flow density ρ to dynamically adjust the dynamic risk factor; S43, Time decay correction: Apply time decay constraints to sudden risk events to generate the final risk value R final ; S44, Risk level classification: Based on the generated final risk value R final , classify the risk levels, including low risk, medium risk, high risk, and critical risk.

8. The vehicle safety management method based on an intelligent parking lot according to claim 7, characterized in that, The risk level division in S44 includes: S441, Low risk: When R final ≤ 50%, it indicates that the current scenario of the parking lot is of low risk; S442, Medium risk: When 50% < R final ≤ 120%, it indicates that the current scenario of the parking lot is of medium risk; S443, High risk: When 120% < R final ≤ 250%, it indicates that the current scenario of the parking lot is medium risk; S444, Critical Risk: When R final > 250%, it indicates that the current scenario of the parking lot is a critical risk.

9. The vehicle safety management method based on an intelligent parking lot according to claim 8, characterized in that, The safety policy trigger in S5 includes: S51, Low-risk policy: When the risk level is low risk, send a voice prompt to the in-vehicle terminal to remind the driver to pay attention, and execute it using the in-vehicle terminal of the target vehicle; S52, Medium-risk policy: When the risk level is medium risk, activate the sound and light warning devices within 3 meters around, and enhance the warning effect through the intelligent roadside unit; S53, High-risk policy: When the risk level is high risk, generate an obstacle avoidance path, and control the intelligent ground lock to rise through the AR navigation system; S54, Critical risk policy: When the risk level is high risk, link the charging piles in adjacent parking spaces to cut off the power supply, and start the physical isolation device, including hydraulic isolation piers for physical safety isolation.

Images