Parking lot management method, device and system for vehicle entrance and exit authenticity verification

Abstract

The invention discloses a parking lot management method, device and system for vehicle entrance and exit authenticity verification, and particularly relates to the technical field of intelligent parking lot management. When a vehicle motion track is reconstructed, a peripheral interference vector and a behavior discontinuous mark are coupled to form an extended track data set; adjacent vehicle interference signals are stripped through principal component analysis to obtain a pure vector, and a Markov chain and gradient descent optimization are integrated to dynamically generate and calibrate an intention branch path; performing weighted fusion correction on the non-standard behaviors, and introducing Monte Carlo avoidance to simulate and predict an interference amplification effect so as to isolate unfinished behaviors; an isolation condition threshold value and a dynamic cooperation threshold value are adaptively adjusted by using the predicted interference amplification effect probability, an access record is generated through a dual threshold value judgment mechanism, and system parameters are iteratively optimized through association rule mining, so that the problems of misjudgment and delay linkage caused by multi-vehicle interference and non-standard behaviors in a high-density parking lot are solved.

Description

Technical Field

[0001] This invention relates to the field of intelligent parking management technology, and more specifically, to a parking management method, device, and system for verifying the authenticity of vehicle entry and exit. Background Technology

[0002] As urban transportation hubs, parking lots are crucial for optimizing space utilization and user experience by verifying the authenticity of vehicle entry and exit. In compact layout scenarios, multi-sensor fusion is needed to reconstruct movement trajectories and determine intentions, ensuring that records are generated only when vehicles completely cross the boundary, thereby preventing billing omissions caused by false entry and exit.

[0003] With the diversification of vehicle types and increased traffic flow, the uncertainty of driver behavior and surrounding interference have become common challenges. For example, temporary reversing or hesitation often occurs due to yielding, while the crossover of signals from neighboring vehicles can confuse recognition, making it difficult for the system to accurately isolate individual behaviors. Vehicle entry and exit verification systems are further vulnerable to judging non-standard driver behaviors such as hesitant reversing intentions due to the crossover of signals from multiple vehicles, leading to a loss of recognition specificity. This is because traditional trajectory reconstruction relies solely on velocity direction vectors, but in reality, when behaviors are discontinuous, they are easily affected by neighboring vehicles, and the mixture of weak signals from electric vehicles and traditional vehicles further exacerbates the confusion.

[0004] Specifically, this complex problem arises from the interference of overlapping contour point clouds with the intention vector, making it impossible to accurately capture physical changes such as positional jitter and the direction of adjacent vehicles. This ultimately leads to collective misjudgment, conflating incomplete actions with interference, causing a chain reaction of recording misalignment, capacity distortion, and delays. Especially during peak hours, this effect amplifies management chaos and safety risks, reduces the system's adaptability to behavioral variability, and if not addressed, will hinder the expansion to high-density intelligent applications, increase intervention requirements, and impact overall efficiency. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, the present invention provides a parking management method, device, and system for verifying the authenticity of vehicle entry and exit. By coupling and purifying the interference vector as a behavioral extension, and utilizing the interference amplification effect probability-driven adaptive adjustment of dual thresholds and game theory equilibrium constraints, the present invention achieves specific and robust identification of vehicle entry and exit intentions under complex interference environments, thereby solving the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a parking lot management method for verifying the authenticity of vehicle entry and exit, comprising: Step S1: In the process of reconstructing the vehicle's motion trajectory, the vehicle's surrounding perception data is jointly processed through spatial grid division and temporal analysis mechanisms to capture interference vectors caused by nearby vehicles or the environment and behavioral discontinuity markers reflecting abnormal vehicle behavior, forming an extended trajectory dataset containing target vehicle trajectory information, surrounding interference vectors, and behavioral discontinuity markers. The interference vectors include at least contour point cloud overlap information, and the behavioral discontinuity markers include at least position jitter or direction reversal information. Step S2: Based on the extended trajectory dataset, the interference vector is treated as a component of the behavior extension and subjected to vector interference stripping. The interference subspace is reduced in dimension to separate the neighboring vehicle ripple signal and extract the pure vector that only reflects the motion characteristics of the target vehicle itself. After obtaining the clean vector, the behavior discontinuity flag is combined with the behavior state transition probability model to perform probabilistic simulation of the discontinuous behavior transitions that the vehicle may have, and dynamically generate multiple intention branch paths with different probability weights; by performing distance-based weight optimization on the intention branch paths and the real-time clean vector, and by introducing the behavior discontinuity flag to provide feedback correction on the path probabilities, the signal stripping process and the intention generation process form a mutually constrained chain purification relationship. Step S3: For the intent branch path, the intent vectors corresponding to multiple intent branch paths are corrected by branch probability fusion, and an avoidance simulation operation is introduced when judging the vehicle entry and exit intent. The behavioral deviation that may be caused by the interference of neighboring vehicles in the multi-vehicle interaction scenario is simulated and predicted to obtain the interference amplification effect probability. Based on the interference amplification effect probability, the incomplete entry and exit behavior is isolated, and the corrected intent vector is output. Step S4: Based on the corrected intent vector, firstly, the isolation conditions used to determine the completion of a single intent are adaptively adjusted according to the probability of the interference amplification effect, and then the corrected intent vector is filtered using the adjusted isolation conditions to obtain effective intent vectors that meet the conditions for completing entry and exit. Subsequently, the proportion of the number of effective intent vectors in all corrected intent vectors is calculated, and the proportion is compared with a dynamic collaboration threshold. Only when the proportion reaches or exceeds the dynamic collaboration threshold is the corresponding vehicle entry or exit record generated to prevent collective misjudgment caused by local interference or a small number of abnormal vectors.

[0007] Preferably, in the spatial grid division, the lane space is divided into uniform three-dimensional grid units, and the side length of each unit is set to a fixed value according to the radar resolution. The ripple signal of the neighboring vehicle is quantified by the phase difference calculation formula. The phase shift of the point cloud data in the grid unit is compared to identify the overlapping parts. These overlaps are regarded as the behavioral extension components and coupled with the behavioral discontinuity marker to form a unified extension set structure.

[0008] Preferably, the location data is subjected to temporal analysis to calculate the difference in position coordinates between consecutive frames, and combined with the direction vector to determine jitter or reversal phenomena, forming a behavior discontinuity mark. The interference vector is connected with the behavior discontinuity mark through grid temporal coupling calculation, and the position jitter or direction reversal amplitude of discontinuous points is marked by the vector norm formula, avoiding inefficient identification caused by fragmented independent processing.

[0009] Preferably, principal component analysis is used to reduce the dimensionality of the interference subspace. By calculating the covariance matrix of the overlapping interference vectors such as contour point clouds in the dataset, the eigenvalues ​​are obtained and sorted. The eigenvectors whose cumulative contribution rate of the eigenvalues ​​reaches a preset ratio are selected as principal components. The original interference vectors are projected onto these principal components to extract pure vectors. The elements of the covariance matrix are the calculation results of the variance and covariance between each dimension of the interference vector. The eigenvalues ​​are obtained by matrix diagonalization. The projection operation is implemented by matrix multiplication.

[0010] Preferably, the process of generating intention branch paths and optimizing their weights and performing feedback correction includes: based on the extracted clean vector and behavior discontinuity markers, using a Markov chain to simulate discontinuous behavior transitions and dynamically generating multiple intention branch paths; simultaneously, by comparing the Euclidean distance between the clean vector and each branch path, using a gradient descent method to iteratively optimize the connection weights of the branch paths to minimize distance errors; and using the magnitude of the behavior discontinuity markers as a feedback factor to correct the confidence of the corresponding state transitions in real time, thereby forming a complementary cleansing chain that integrates signal stripping, path generation, and feedback correction.

[0011] Preferably, correcting non-standard behavior and predicting interference amplification effects includes: weighting and fusing the corresponding vectors of multiple intent branch paths using branch probabilities to form a single calibration intent vector; using Monte Carlo sampling, repeatedly and randomly adding neighboring vehicle interference variables following a preset distribution to the calibration intent vector to generate multiple simulated trajectories; in each sampling, a dynamic boundary threshold (used for judging whether the simulated trajectory exceeds the boundary) is determined by the sum of the basic boundary threshold and the probability of the preceding interference amplification effect multiplied by the dynamic offset coefficient; statistically analyzing the proportion of all samples where the deviation exceeds the dynamic boundary threshold as the current probability of interference amplification effect; when the probability of interference amplification effect is higher than the preset boundary value, the corresponding intent vector is marked as an incomplete behavior and removed.

[0012] Preferably, the intention vector is calibrated using a branch probability weighted fusion operation. By calculating the probability value of each branch path as a weight, the corresponding intention vectors in the path are weighted and summed. The weighted summation formula is that the corrected intention vector is equal to the sum of the products of each branch intention vector and its corresponding probability. This fuses multiple branches into a single calibrated intention vector. The intention vector weight of the low-probability branch is multiplied by the reciprocal of the interference amplification effect probability for penalty adjustment. The reciprocal is calculated as 1 divided by the interference amplification effect probability. When the interference amplification effect probability is close to 1, the penalty magnitude increases, thereby correcting the confusion of non-standard behavior. (By calibrating multiple branches into a single vector through probability weighted fusion, the problem of intention ambiguity caused by discontinuous behavior in crowded parking lots is solved.)

[0013] Preferably, adjusting the isolation condition of the intent vector based on the predicted effect refers to: based on the interference amplification effect probability calculated in step S3, subtracting it from the preset interference effect benchmark value, and then performing nonlinear mapping through the hyperbolic tangent saturation function to obtain the adjustment coefficient; multiplying the adjustment coefficient by the preset isolation threshold amplification coefficient, adding it to 1, and then multiplying it by the original isolation threshold to obtain the adjusted isolation condition threshold; and then using the adjusted isolation condition threshold to perform geometric distance screening on the corrected intent vector to obtain the effective intent vector.

[0014] Preferably, the game theory equilibrium optimization in step S4 includes the following steps: A two-player game model is established between threshold adjustment strategy and interference environment state. The pure strategies of threshold adjustment strategy include relaxing the isolation threshold and tightening the isolation threshold, and the pure strategies of interference environment state include high interference scenario and low interference scenario. The revenue value corresponding to each strategy combination in the revenue matrix is ​​determined by statistical analysis of historical data or online learning. The revenue value is calculated by combining the recognition accuracy, false alarm rate and passage delay time according to preset weights. The Nash equilibrium point of the payoff matrix is ​​solved by iteratively eliminating inferior strategies to obtain the corresponding equilibrium solution. The equilibrium solution is mapped to the theoretical upper limit of the isolation condition threshold adjustment. During operation, if the threshold adjustment suggestion calculated based on the real-time interference amplification effect probability exceeds the theoretical upper limit, the theoretical upper limit is used as the actual adjustment result.

[0015] Preferably, the method further includes: Step S5: Upload the recorded results to the cloud, and feed back the isolated parameters through cross-event feedback to spatial grid partitioning and temporal analysis to achieve incremental optimization of composite behavior; The isolation parameters include an isolation condition threshold used in step S4 to filter valid entry and exit intentions from the correction intention vectors, and a dynamic cooperation threshold used to determine the overall consistency of the set of valid intention vectors. The isolation condition threshold is a geometric distance parameter used to compare with the distance value from the correction intention vector to the preset virtual lane boundary to determine whether the intention of a single vehicle meets the spatial location requirements for completing entry and exit. The dynamic cooperation threshold is a proportional parameter used to compare with the proportion of the number of valid intention vectors to the total number of correction intention vectors to determine whether the overall consistency level of the intention meets the requirements for generating a complete entry and exit record. Both the isolation condition threshold and the dynamic cooperation threshold can be dynamically adjusted based on the interference amplification effect probability calculated in the previous steps, and iteratively optimized in step S5 through a cloud-based cross-event feedback mechanism to achieve adaptation to the complex interference environment.

[0016] The technical effects and advantages of this invention are as follows: (1) This invention couples multi-vehicle interference signals (such as overlapping contour point clouds) with discontinuous signs of target vehicle behavior (such as position jitter) into a unified extended trajectory dataset, and treats it as a separable and predictable extended behavioral component in subsequent steps, rather than simply filtering independent noise, thus achieving a fundamental change in the interference processing method. Furthermore, by constructing a chain-like processing flow from signal stripping, intent generation, risk simulation to dynamic decision-making, and embedding closed-loop connection mechanisms such as gradient descent optimization, sign feedback correction, and probability penalty adjustment in each link, the front-end signal purification and back-end behavior prediction are deeply coordinated, effectively solving the technical problem of loss of vehicle recognition specificity and high misjudgment rate of non-standard behavior caused by fragmented processing in high-density, high-interference parking lot scenarios in traditional methods.

[0017] (2) This invention further introduces a Monte Carlo sampling-based interference amplification effect quantification prediction, providing a real-time risk perception signal for subsequent record generation; and adopts a dual adaptive judgment architecture driven by the probability of this effect, namely, an isolation condition threshold and a dynamic cooperation threshold. The isolation threshold is adjusted by a nonlinear saturation function to achieve robust relaxation or tightening of geometric judgment, while the cooperation threshold is constrained by game theory equilibrium to maintain a stable upper limit for overall consistency requirements. The two work together to achieve a shift from precise single-vector screening to overall consistency. Figure 1 The two-layer decision insurance with consistency confirmation effectively solves the problem of collective misjudgment and chain reaction of delays caused by cross-interference of multiple vehicles, and significantly improves the accuracy and robustness of access management in parking lots with compact layouts during peak hours. Attached Figure Description

[0018] Figure 1 This is a simplified diagram of the parking lot management process for verifying the authenticity of vehicle entry and exit, as described in this invention.

[0019] Figure 2 This is a flowchart illustrating the process of obtaining the intention branch path in this invention. Detailed Implementation

[0020] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0021] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0022] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the scope of this application and its application or use.

[0023] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0024] Example 1 addresses the issue of "false recording" caused by vehicles making U-turns or turning around in parking lot entrance and exit areas, ensuring that billing records match the actual parking status. This example provides a method for verifying the authenticity of vehicle entry and exit, applicable to parking lot entrance and exit lanes. Based on the collaborative operation of multiple sensors such as inductive loops, vehicle-passing radar, and auxiliary cameras, a rigorous decision-making process is established to ensure that a valid record is generated only when a vehicle actually completes its entry or exit behavior, thus avoiding billing errors caused by misjudgment. The method includes the following steps: First, multi-sensor collaborative triggering and synchronous data acquisition: Using a ground loop as the primary trigger source, the physical presence of a vehicle is detected. Once the loop is triggered, the vehicle-passing radar and auxiliary camera are immediately activated to simultaneously acquire the vehicle's speed, distance, contour point cloud data, and visual images. Through the built-in spatiotemporal alignment algorithm, the multi-source data is calibrated and fused in time and space to provide a consistent and reliable data foundation for subsequent analysis. Then, the vehicle's trajectory is reconstructed and its entry / exit intention is accurately determined: Based on the fused sensor data, the vehicle's continuous trajectory within the lane is reconstructed in real time, and its direction and speed are analyzed. By using preset virtual lane boundary lines (such as key positions behind the barrier gate), it is determined whether the vehicle has completed the key behavior of "crossing the boundary and continuously driving away". If the trajectory completely crosses the entry line from the outside to the inside, it is determined to be a real entry; if it completely crosses the exit line from the inside to the outside, it is determined to be a real exit; otherwise, it is considered as not completing the entry or exit. Finally, intelligent event decision-making and system fault tolerance: Based on the combined license plate recognition results and trajectory judgment conclusions, a final decision is made: if recognition is successful and the trajectory judgment indicates a genuine entry or exit, a complete record is generated; if recognition is successful but no genuine entry or exit is detected, the event is ignored and an anomaly is recorded; if recognition fails but the trajectory judgment indicates a genuine entry or exit, an anonymous record is generated and an auxiliary review process is initiated. Simultaneously, the status of each sensor is continuously monitored, and when a single sensor failure is detected, the system automatically switches to a degraded operation mode and adjusts the confidence threshold to ensure reliable operation even under non-ideal conditions.

[0025] Example 2 is based on Example 1. In order to solve the problems of fragile non-standard behavior judgment and multi-vehicle interference in Example 1, it realizes intention prediction by extending trajectory isolation through communication and realizes dynamic collaboration through cloud cross-event feedback. For details, please refer to Figure 1 A simplified flowchart of the parking lot management process for verifying the authenticity of vehicle entry and exit is provided in Embodiment 2 of this invention. The parking lot management method for verifying the authenticity of vehicle entry and exit includes the following steps: Step S1: When reconstructing the vehicle's trajectory, the surrounding interference vectors and behavioral discontinuity markers are captured through spatial grid division and temporal analysis mechanisms to form an extended trajectory dataset. The interference vectors include overlapping contour point clouds, and the behavioral discontinuity markers include position jitter or direction reversal. Overview: For step S1, the inputs are the raw signals of vehicle motion, such as velocity direction vectors, contour point clouds, and position data. The output is an extended trajectory dataset containing surrounding interference vectors (such as point cloud overlap) and behavior discontinuity markers (such as jitter reversal). This is achieved by using spatial grid partitioning and temporal analysis mechanisms to couple interference and behavior into an extended set. That is, starting from the beginning by calculating the adjacent vehicle waves and marking discontinuities with the parallel vector norm, to the end by providing a complete chain of accurate basic inputs, interference is treated as behavior extension rather than noise filtering. Through complex coupling and connection of grid and temporal sequences, a chain-like unification of fragmented processing is achieved, avoiding the inefficiency of traditional independent processing. Furthermore, the detailed implementation of step S1 includes: Step 101: Use vehicle radar to collect vehicle velocity direction vector and contour point cloud data, and auxiliary camera to collect vehicle position data. Use these data as raw input, and partition and quantize the contour point cloud data through spatial grid partitioning method. Divide the lane space into uniform three-dimensional grid units. The side length of each unit is set to a fixed value according to the radar resolution to capture the overlap of contour point cloud as the surrounding interference vector. At the same time, perform time series analysis on the position data, calculate the position coordinate difference between consecutive frames, and combine it with the direction vector to judge the jitter or reversal phenomenon, forming behavior discontinuity mark, thereby constructing an extended trajectory dataset containing interference vector and behavior discontinuity mark. Step 102: Based on the interference vector in the extended trajectory dataset, the ripple signal of the nearby vehicle is quantized by the phase difference. Specifically, the overlapping parts are identified by comparing the phase shift of the point cloud data in the grid cell, and these overlaps are regarded as the behavioral extension components. Coupled with the behavioral discontinuity marker, a unified extended set structure is formed to ensure that the interference signal is not regarded as independent noise but is integrated into the trajectory analysis chain. Step 103: Perform chain-like unified processing on the extended set. Connect the interference vector with the behavior discontinuity flag through grid temporal coupling calculation. Specifically, use the vector norm formula to mark the position jitter or direction reversal amplitude of the discontinuity point, and use these flags as basic inputs and outputs to the subsequent separation steps.

[0026] Step S2: Based on the extended trajectory dataset, use the vector interference stripping algorithm to separate the neighboring vehicle ripple signals and integrate the behavior probability calculation to dynamically generate the intent branch path; Overview: For step S2, the input is the extended trajectory dataset of S1, and the output is the intent branch path. The means of implementation is to separate the neighboring vehicle signals and dynamically construct the path by combining the vector interference stripping algorithm with the dimensionality reduction generation mechanism of behavior probability calculation. That is, it starts with extracting pure vectors from the dimensionality reduction interference subspace of principal component analysis. The probability calculation uses Markov chain to simulate discontinuous transitions to ensure the identification specificity of a single vehicle. Stripping and generation are regarded as complementary purification chains. The complex connection is achieved by dynamically calculating the subspace, so as to realize the chain-like intelligent purification of confused signals rather than static removal. Furthermore, the vector interference stripping algorithm first calculates the covariance matrix of interference vectors (such as overlapping contour point clouds) in the extended trajectory dataset, and extracts eigenvalues ​​and eigenvectors through matrix diagonalization; then, it selects the principal eigenvectors with a cumulative contribution rate exceeding a preset proportion to construct a pure subspace, projects the original high-dimensional interference vectors onto this subspace, thereby stripping away the neighboring vehicle's ripple signal and obtaining a pure vector that only reflects the target vehicle's own motion characteristics; the integrated behavior probability calculation refers to, after separating the pure vector, using the pure vector representing the vehicle's current motion state and the behavior discontinuity marker as initial inputs, and querying the state transition probability matrix constructed based on massive historical trajectory data statistics. It simulates the probability of a vehicle evolving from its current state to the next possible behavioral state (such as going straight, turning left, turning right, reversing, stopping, shaking, or reversing). Through chain iteration and matrix multiplication, it dynamically deduces multiple future intention branch paths with different probability weights, forming a behavior prediction tree covering multiple possibilities. This process is not static matching, but rather continuously optimizes the Euclidean distance between the predicted path and the real-time pure vector through gradient descent, and embeds feedback of behavior discontinuity markers to correct the probability confidence in real time. This achieves closed-loop connection and adaptive calibration between the behavior probability and the stripped signal, providing quantitative and multi-dimensional intent inference basis for core decision-making.

[0027] In one possible embodiment, see [reference] Figure 2 The flowchart for obtaining the intention branch path is shown in the figure. The detailed implementation of step S2 includes: Step 201: Dimensionality reduction of the interference subspace and extraction of pure vectors: In the process of separating neighboring vehicle signals based on extended trajectory datasets, principal component analysis is used to reduce the dimensionality of the interference subspace. By calculating the covariance matrix of the overlapping interference vectors such as contour point clouds in the dataset, the eigenvalues ​​are obtained and sorted. The eigenvectors whose cumulative contribution rate of eigenvalues ​​reaches a preset ratio are selected as principal components (usually a preset ratio of 85%-95%, which can be adjusted according to radar accuracy). The original interference vectors are projected onto these principal components to extract pure vectors. The elements of the covariance matrix are defined as the calculation results of the variance and covariance between each dimension of the interference vector. The eigenvalues ​​are obtained by matrix diagonalization. The projection operation uses matrix multiplication to extract pure vectors, thereby achieving the initial stripping of interference signals. Step 202: Behavioral probability modeling and intent branch generation based on Markov chains: Based on the extracted pure vectors, the behavior probability calculation operation is integrated, and a Markov chain is used to simulate discontinuous behavior transitions to construct a state transition probability matrix. The matrix rows represent the current behavior state, such as position jitter or direction reversal, and the columns represent the next state. The element values ​​are the transition probabilities obtained from historical trajectory data (the statistical method is count normalization). The probability path from the initial state to the target state is calculated through chain iteration. The initial state is defined by the behavior discontinuity flag in the pure vector. The iteration uses matrix multiplication to accumulate probabilities, thereby dynamically generating intention branch paths and ensuring that the separated signals and behavior probabilities are complementary and connected. Step 203: Dynamic weighting of the pure vector and the intent path: The generated intention branch path is subjected to subspace dynamic computation to complicate the connection operation. The connection weight is adjusted by comparing the pure vector with the Euclidean distance of the branch path. The Euclidean distance is defined as the square root of the sum of the squares of the differences in the corresponding dimensions of the vectors. The weight adjustment adopts the gradient descent method to minimize the distance error. The gradient calculation is based on the partial derivative of the distance with respect to the weight. The weight is iteratively updated until convergence. Thus, the stripping operation and path generation are regarded as complementary purification chains, realizing chain-like intelligent purification of confusing signals and avoiding the loss of specificity caused by static removal. Step 204 involves probability confidence correction driven by discontinuous flags: The feedback logic of behavior discontinuity flags is embedded in the cleanup chain. Specifically, after calculating the probability value of each intention branch path using a fixed state transition probability matrix, the probability of the relevant path is temporarily corrected according to the amplitude of the behavior discontinuity flag. When the flag indicates a fluctuating position, the probability value of the corresponding state transition is multiplied by a decay factor based on the flag amplitude, thereby reducing the reliability of making deterministic predictions under abnormal behavior states. The corrected probability is only used for path selection at the current moment and does not change the fixed state transition probability matrix itself, thus ensuring the temporary adaptability of path construction to non-standard behaviors. Step 205: High-confidence intent branch path filtering and output: The output is the optimized intent branch path as the final result. The path includes the probability sequence and corresponding vector of multiple branches to ensure that the specificity of a single vehicle recognition is filtered by the path probability threshold. (The path probability threshold is used to filter the intent branch paths generated by the Markov chain, filtering out low-probability noise paths. The filtered high-confidence paths are used as the input for step S3. The path probability threshold can be set to a fixed proportion based on historical verification data. A PR curve is plotted using historical data, and the threshold corresponding to the point with the maximum F1 score is selected. The example value is 0.65.) This completes the entire chain from interference removal to intent generation.

[0028] To illustrate this, consider the following example: Assume there's a vehicle trajectory in reality. Due to interference from neighboring vehicles, the contour point clouds overlap significantly, and the vehicles themselves are jittering. After PCA in step 201, a low-dimensional clean vector is obtained, which removes most of the overlapping interference but retains the jitter features. In step 202, the Markov chain generates multiple possible paths based on the jitter features (initial state): one is a normal exit (probability 0.6); one is reversing (probability 0.3); and one is hesitant stopping (probability 0.1). In step 203, the Euclidean distance between the clean vector and each path is calculated. It is found that the normal exit path has the smallest distance, but the reversing path also has a relatively small distance. By adjusting the weights through gradient descent (for example, setting the parameters to: learning rate η=0.01, maximum number of iterations 1000, and convergence condition of distance change <0.001), the probability weight of the reversing path may be reduced, while the weight of the normal exit is increased. In step 204, due to the large amplitude of the jitter flag, the confidence of the transition probability from the jitter state to the normal exit is reduced, thereby further adjusting the path probability. Finally, in step 205, if the path probability threshold is set to 0.5, then the normal exit path (adjusted probability 0.7) is retained, the reversing path (adjusted probability 0.25) is filtered out, and the final output intention is normal exit.

[0029] The explanation explains that the path probability threshold is used to filter out high-confidence effective paths from multiple generated intent branch paths for subsequent processing. It ensures the specificity of individual vehicle identification by filtering low-probability noise paths. In addition to traditional empirical setting or simple statistical quantile methods, the threshold can also be selectively obtained using a scene adaptive dynamic calibration mechanism based on reinforcement learning. This embodiment of the invention does not specifically limit this. The adaptive dynamic calibration mechanism uses real-time recognition accuracy and false alarm rate as reward signals to construct a Markov decision process model. It automatically explores and optimizes the threshold parameters through online policy gradient learning, enabling it to autonomously adjust in different parking scenarios (such as peak hours, off-peak hours, and charging areas) and different time periods. At the same time, the initialization stage integrates the probability distribution of correct and incorrect cases in historical verification data for cross-validation and establishes a correlation feedback between the threshold and behavior discontinuity indicators.

[0030] Step S3: Based on the generated intent branch path, correct non-standard behaviors, introduce avoidance simulation operation when judging the entry and exit intent, predict the interference amplification effect to isolate the incomplete behavior, and output the corrected intent vector. The explanation is as follows: Avoidance simulation operation refers to the process of predicting the specific handling operations for interference amplification effects by using Monte Carlo sampling to randomly simulate the avoidance disturbances that may be caused by adjacent vehicles during the process of judging vehicle entry and exit intentions. This operation uses the calibration intention vector as the basic input, adds interference variables to the intention vector to generate multiple simulated trajectories, and then calculates the deviation of each simulated trajectory from the virtual lane boundary. The virtual lane boundary can be a lateral baseline 1.5 meters behind the barrier gate, with coordinates determined by the parking lot CAD drawings; for example, with the center of the barrier gate as the origin, the lane direction as the X-axis, and the boundary line at X=1.5m (1.5 meters behind the barrier gate), the proportion of deviations exceeding the boundary threshold is statistically analyzed as the probability of interference amplification effects. This enables a predictive amplification effect assessment of non-standard behaviors such as hesitation, reversing, or temporary stopping caused by drivers avoiding adjacent vehicles in actual parking lots. It can more accurately isolate incomplete entry and exit behaviors in multi-vehicle signal intersection environments, avoiding misjudging brief movements affected by avoidance as genuine entry and exit intentions. Among them, the interfering variables, such as the position offset and speed disturbance of the adjacent vehicle, are assumed to follow a normal distribution N(0, σ²), where σ is obtained from historical avoidance data. The boundary threshold refers to the geometric distance standard used in Monte Carlo sampling to determine whether the deviation between the simulated trajectory and the virtual lane boundary exceeds the limit; it is used to calculate the probability of interference amplification effect and is determined as a fixed value based on historical verification trajectory statistics. The Monte Carlo sampling method works by simulating the amplification of interference caused by adjacent vehicles avoiding each other in a real parking lot through random perturbation. The input is the calibration intention vector and the distribution of the interference variables, and the output is the probability of the interference amplification effect. The intrinsic connection is that the sampling trajectory deviation directly reflects the amplification risk of non-standard behavior in a multi-vehicle environment. When combined with the compact layout of parking lots, it can effectively isolate hesitation or reversing behavior and avoid collective misjudgment.

[0031] Overview: For step S3, the input is the intent branch path of S2, and the output is the corrected intent data. The means of implementation is to use a probability-weighted fusion mechanism of non-standard behavior confusion correction combined with avoidance simulation operation to calibrate the path and predict the amplification effect. That is, starting from the beginning with the branch probability weighted fusion intent vector, the complete chain of incomplete behavior is isolated at the end. The fusion correction and simulation are the effect chain. Through Monte Carlo sampling of complex random logic, the chain of accurate isolation of the preceding path-driven judgment is ensured.

[0032] Furthermore, the detailed implementation of step S3 includes: Step 301: In the process of correcting non-standard behavior confusion in intent branch paths, the intent vector is calibrated by using a branch probability weighted fusion operation. Specifically, the probability value of each branch path is calculated as a weight, and the corresponding intent vectors in the path are weighted and summed. The probability value comes from the transition probability generated by the preceding Markov chain. The intent vector consists of a pure vector and a branch sequence. The weighted summation formula is that the corrected intent vector is equal to the sum of the products of each branch intent vector and its corresponding probability. Thus, multiple branches are fused to form a single calibrated intent vector, achieving the initial correction of the confused part. Step 302: Based on the calibration intent vector, an avoidance simulation operation is introduced. Monte Carlo sampling is used to predict the interference amplification effect. Specifically, multiple simulated trajectories are generated by repeatedly sampling the interference variables. The interference variables are defined as the position offset and speed disturbance of the adjacent vehicle. Each time the interference variables are sampled, they are added to the calibration intent vector to calculate the simulated position sequence. The number of samplings is set to a fixed value to cover the probability distribution. The deviation of each simulated trajectory from the virtual lane boundary is calculated. The proportion of deviations exceeding the boundary threshold is used as the probability of interference amplification effect. Step 303: Isolate incomplete behaviors based on the probability of interference amplification effect; specifically, when the probability of interference amplification effect is higher than the preset boundary value, mark the corresponding vector as an incomplete behavior and remove it from the calibration intent vector; the preset boundary value is set to a fixed proportion based on historical verification trajectory statistics; after removal, the weights of the remaining vectors are renormalized to ensure that the calibration intent data only retains the vectors corresponding to complete entry and exit behaviors; Step 304: Embed the constraint feedback logic of branch probability in the effect chain. Specifically, the intention vector weight of the low probability branch is multiplied by the reciprocal of the interference amplification effect probability for penalty adjustment. The reciprocal is calculated as 1 divided by the interference amplification effect probability. When the interference amplification effect probability is close to 1, the penalty magnitude is increased, thereby strengthening the chain's sensitivity to interference amplification and ensuring the accurate isolation of the direct driving judgment of the preceding branch path. Step 305: Output the isolated correction intent vector as the final result. The data includes the correction intent vector and the probability sequence of the remaining branches, ensuring that incomplete behaviors have been eliminated, thus completing the entire chain from path calibration to effect prediction.

[0033] For ease of understanding, the following formula for calculating the probability of interference amplification effect is provided in the embodiments of the present invention: ; in, It represents the probability of interference amplification effect, with a value range of [0, 1]. Its function is to quantify the risk ratio of intention deviation amplification under multi-vehicle interference, and to serve as a feedback driving signal for threshold adjustment. Indicates the first The simulated trajectory deviation magnitude of the Monte Carlo sampling is calculated by Euclidean distance and takes the value of a non-negative real number. It is used to characterize the degree of deviation between the intention vector and the lane boundary under a single disturbance. This indicates an indicator function, taking the value 0 or 1, which counts the number of samples exceeding the dynamic boundary threshold. This represents the basic boundary threshold, with a value range of positive real numbers (based on lane geometry fixation), and serves as a static isolation benchmark. This represents the dynamic offset coefficient, which takes the value of a positive real number (fixed empirically), and its function is to amplify the boundary sensitivity based on the preceding effect. This represents the probability of the amplification effect of the preceding interference, with a value range of [0, 1]. Its function is to introduce temporal memory and avoid single fluctuations. This represents the total number of samples, and its value ranges from positive integers. Its purpose is to ensure statistical stability.

[0034] Step S4: Perform a record generation operation using the corrected intent vectors, adjust the isolation conditions of the intent vectors according to the prediction effect, and filter out the effective intent vectors from the intent vectors; calculate the proportion of the number of effective intent vectors in all corrected intent vectors, and compare it with the dynamic collaboration threshold to prevent collective misjudgment from causing a chain reaction of delays. The explanation explains that step S4 employs a dual threshold determination mechanism to ensure the authenticity of vehicle entry and exit records. First, the isolation condition threshold is adaptively adjusted based on the probability of interference amplification: when interference is significant, the geometric distance requirement for individual vector determination is appropriately relaxed; conversely, it is tightened. Second, the adjusted isolation condition threshold is used to filter each corrected intent vector, retaining vectors that meet the distance condition as valid intent vectors. Finally, the proportion of valid vectors in the total vectors is calculated and compared with the dynamic cooperation threshold. If the proportion reaches or exceeds the dynamic cooperation threshold, the overall intent is considered valid. Figure 1 If the interference is consistent, a complete entry / exit record is generated; otherwise, an abnormal ignore record is generated. Both thresholds can be dynamically adjusted according to the real-time interference level. The isolation condition threshold is mainly corrected linearly based on the deviation between the interference amplification effect probability and the baseline value, while the dynamic cooperation threshold may be adjusted in the opposite direction (i.e., the consistency requirement is reduced when the interference is large and increased when the interference is small), thus forming a cooperative adaptive mechanism.

[0035] Furthermore, the dynamic collaboration threshold is set as follows: Step (1) Baseline initialization: Based on the offline historical dataset, the proportion distribution of effective intent vectors in real vehicle entry and exit events is statistically analyzed, and the preset percentile is used as the initial baseline value of the dynamic collaboration threshold, thereby establishing the basic level of consistency judgment. Step (2) Dynamic feedback adjustment: During operation, the real-time interference amplification effect probability is received from the preceding processing stage. Using the aforementioned probability as a feedback factor, the dynamic cooperation threshold is adaptively optimized online through a predefined weighted correction function or saturation adjustment function, so that the consistency judgment requirement is relaxed when the interference amplification effect is significant, and the requirement is increased otherwise. Step (3) Equilibrium Constraints to Prevent Drift: By modeling with game theory, the Nash equilibrium point of vehicle intention and interference effect under strategy game is solved, and the threshold corresponding to the equilibrium point is used as the theoretical upper limit of dynamic adjustment; when the threshold obtained by feedback adjustment in step (2) exceeds the upper limit, the upper limit value is used as the final dynamic cooperation threshold to avoid excessive threshold drift in continuous or extreme interference scenarios. The coefficients in the weighted correction function, the parameters of the saturation adjustment function, and the initial benchmark value are all determined and solidified through offline historical data training and optimization before system deployment, thus forming a complete adaptive closed loop with benchmark reference, real-time feedback, and theoretical constraints, so as to achieve sensitive response to complex interference environments while ensuring decision stability.

[0036] For ease of understanding, the following calculation formula for the isolation condition threshold is provided in this embodiment of the invention: First, the isolation condition threshold is adaptively adjusted using the following formula: ; in, The adaptive adjustment value representing the isolation condition threshold is a dynamic geometric threshold used to compare the distance between the vector and the virtual lane boundary under the current event. The value range is a positive real number (fluctuating around the original value). This represents the original threshold for isolation conditions, determined by statistical analysis of lane geometry and historical complete entry and exit events using an offline verification dataset. The value range is a positive real number (with a fixed baseline, such as a certain proportion of the lane width). Its function is to serve as the starting baseline for adaptive adjustment, ensuring that the threshold does not deviate from its physical meaning. This represents the isolation threshold amplification factor, which is determined by optimization using the offline dataset and is a fixed positive real number with a value range of [0.1, 1.0]. Its function is to control the sensitivity of the interference effect probability to the threshold and avoid over- or under-adjustment. This represents the hyperbolic tangent saturation function, with a value range of (-1, 1). Its function is to prevent nonlinear mapping interference deviation, achieve smooth saturation of threshold adjustment, and prevent the threshold from being too large or too small when the effect probability is extreme. The probability of interference amplification effect is represented by the Monte Carlo sampling statistics in step S3. The value range is [0, 1]. It acts as a feedback factor to drive the real-time adaptation of the isolation threshold. This represents the baseline value for the interference effect. It is a fixed value determined by statistical normality from the offline dataset, and its value range is [0, 1]. Its function is to define the reference zero point when there is no interference.

[0037] For ease of understanding, the following formula for calculating the dynamic collaboration threshold is provided in this embodiment of the invention: ; The Nash equilibrium value in game theory is used as an upper limit constraint to achieve dynamic optimization of the cooperation decision ratio; This represents the dynamic collaboration threshold, indicating the effective intent used for the final record generation. Figure 1 The consistency ratio threshold has a value range of [0, 1]. This represents the collaboration threshold baseline, which is determined by the offline dataset as a fixed value (e.g., 0.7), with a value range of [0, 1], and serves as the starting point for dynamic adjustment; This represents the collaboration threshold adjustment coefficient, which is determined by offline optimization to be a fixed positive real number with a value range of [0.05, 0.3]. Its function is to control the sensitivity to changes in time-series effects. This represents the probability of the interference amplification effect of the preceding event, with a value range of [0, 1]. Its function is to introduce temporal smoothing and avoid the drastic impact of a single fluctuation on the cooperation threshold. This represents the Nash equilibrium value in game theory, which is obtained by iteratively solving the strategies of both parties (goal intention vs. interference effect). The value range is [0, 1], and it serves as an upper limit constraint on the cooperation threshold to ensure that the adjustment does not exceed the optimal boundary of the adversarial equilibrium. summary: (Step S3 outputs the probability of interference amplification effect) Lateral drive (Geometric screening threshold) and Adaptive calculation of (consistency ratio threshold); First, vector distance comparison is used to determine the valid intent; then, consistency is used to determine the remaining intent. (by The upper limit constraint ultimately determines the type of record result.

[0038] Furthermore, step S4, which utilizes the corrected intent vector to perform the record generation operation, includes the following steps: Step 401: Adaptive adjustment of isolation condition threshold Before the execution record is generated, the isolation condition threshold used to determine whether a single vehicle has completed its entry and exit behavior is adaptively adjusted based on the interference amplification effect probability obtained in step S3. The original isolation threshold is preset to describe the basic geometric boundary requirements required for a vehicle to complete its entry and exit. When the interference amplification effect probability is high, the original isolation threshold is weighted and corrected by introducing the interference amplification effect probability as a feedback factor to form the adjusted isolation condition threshold, so as to avoid false rejection or false retention caused by a fixed threshold in strong interference scenarios. Step 402: Filtering effective intent vectors based on isolation condition thresholds For each corrected intent vector, it is projected into the lane coordinate system, and the vertical Euclidean distance (geometric distance) between the endpoint of the intent vector and the preset virtual lane boundary is calculated and compared with the adjusted isolation condition threshold: when the geometric distance meets the isolation condition threshold requirement, it is determined that the vehicle behavior corresponding to the intent vector has completed the actual entry and exit and is retained as a valid intent vector; when the geometric distance does not meet the isolation condition threshold requirement, it is determined that the behavior is still in the approach, hesitation or incomplete state and the corresponding intent vector is removed. Step 403: Multi-meaning based on dynamic collaboration threshold Figure 1 Confirmation of consistency After completing the isolation and filtering of a single intent vector, a dynamic collaboration threshold is introduced to confirm the consistency of the overall judgment result for the set of retained valid intent vectors. Specifically, the proportion of the number of valid intent vectors in all corrected intent vectors is calculated and compared with the dynamic collaboration threshold. When the proportion reaches or exceeds the dynamic collaboration threshold, it is determined that a consistent entry and exit judgment is formed among multiple valid intent vectors, and a corresponding complete vehicle entry and exit record is generated. When the proportion does not reach the dynamic collaboration threshold, it is determined that the current judgment result lacks overall consistency, and only anomaly ignore records are generated, thereby avoiding false recordings triggered by a small number of residual interference vectors; based on the judgment result of the dynamic collaboration threshold, vehicle entry and exit records or anomaly ignore markers are output as the record generation result of step S4, providing a reliable input basis for cross-event feedback and parameter iteration in subsequent step S5.

[0039] Furthermore, the game theory equilibrium optimization in step S4 includes the following steps: A two-player game model is established between threshold adjustment strategy and interference environment state. The pure strategies of threshold adjustment strategy include relaxing the isolation threshold and tightening the isolation threshold, and the pure strategies of interference environment state include high interference scenario and low interference scenario. The payoff value for each strategy combination in the payoff matrix is ​​determined through historical data statistics or online learning. The payoff value is calculated by combining the recognition accuracy, false alarm rate, and passage delay time according to preset weights. The Nash equilibrium point of the payoff matrix is ​​solved by iteratively eliminating inferior strategies to obtain the corresponding equilibrium solution. The equilibrium solution is mapped to the theoretical upper limit of the isolation condition threshold adjustment. During operation, if the threshold adjustment suggestion value calculated based on the real-time interference amplification effect probability exceeds the theoretical upper limit, the theoretical upper limit is used as the actual adjustment result.

[0040] The explanation is that the record result refers to the final judgment and record output of the current vehicle entry and exit event. This result directly determines whether a valid entry and exit record is generated and subsequent management behaviors such as billing and parking space statistics. Step S5: Upload the recorded results to the cloud, and feed back the isolated parameters through cross-event feedback to spatial grid partitioning and temporal analysis to achieve incremental optimization of composite behavior; For step S5, the input is the recorded result of S4, and the output is the iterative isolation parameter fed back to S1. The means of achieving this is to refine the result into a pattern and optimize the parameter through the cloud-based cross-event feedback association rule mining mechanism. Furthermore, the specific implementation method corresponding to step S5 is as follows: Step 501: During the process of uploading the recorded results to the cloud, the association rule mining operation is used to refine the pattern of multi-vehicle events. For example, the Apriori algorithm is used to scan the set of recorded results to generate frequent itemsets. The itemset is defined as a combination of isolation parameters such as the probability pair of threshold boundary and interference amplification effect. The support is the number of times the itemset appears divided by the total number of records. The confidence is the support of the conditional itemset divided by the support of the antecedent. Strong association rules are extracted as multi-vehicle patterns with a fixed minimum support and confidence threshold. Step 502: Optimize the isolation parameter operation based on the extracted strong correlation rules. Specifically, the consequent parameter value of the rule is used as the candidate optimization target to replace and adjust the current isolation parameter. The rule with the highest confidence is selected for replacement. When multiple rules conflict, the rule with the highest support is selected to ensure that the parameter optimization directly reflects the pattern refinement results of cross-event feedback. Step 503: Perform a feedback operation on the optimized isolation parameters; specifically, transmit the parameters to the adjacent vehicle signal isolation processing stage of the trajectory reconstruction stage through an encrypted channel, directly overwriting the original parameter values, thereby forming a feedforward chain and realizing the system's incremental optimization of composite behaviors; Step 504: In association rule mining, add event feedback constraints, treating the anomaly ignore markers in the recorded results as negative itemsets for scanning. When the support of a negative itemset exceeds a threshold, suppress the generation of the corresponding rule, enhance the sensitivity to misjudged events, and ensure a chain evolution from local isolation to global optimization. The operating principle of association rule mining is to extract cross-event co-occurrence patterns through frequent itemsets and strong rules. The input is a set of record results containing complete records and ignored tags, and the output is optimized isolation parameters. The inherent connection is that the rule consequents directly guide parameter replacement. When combined with the parking lot multi-vehicle peak period scenario, it can effectively iteratively optimize interference isolation and avoid the long-term accumulation of misjudgments of compound behaviors.

[0041] Summary: Based on the multi-sensor collaborative triggering, trajectory reconstruction and decision fault tolerance of Example 1, Example 2 further optimizes the combined effects of behavioral discontinuity and signal intersection. Through vector stripping and avoidance simulation logic, the system specificity is ensured. Interference is regarded as behavioral branches rather than static noise. Through progressive isolation and effect prediction, the complex processing context is made more complex, avoiding the limitations of simple filtering.

[0042] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A parking lot management method for verifying the authenticity of vehicle entry and exit, characterized in that, include: Step S1: When reconstructing the vehicle's trajectory, the surrounding interference vectors and behavioral discontinuity markers are captured through spatial grid division and temporal analysis mechanisms to form an extended trajectory dataset. The interference vectors include overlapping contour point clouds, and the behavioral discontinuity markers include position jitter or direction reversal. Step S2: Based on the extended trajectory dataset, the neighboring vehicle ripple signal is separated from the target vehicle signal using a vector interference stripping algorithm. The interference vector is treated as a component of the behavior extension and principal component analysis is performed to extract the pure vector. The behavior probability calculation is then integrated. Based on the pure vector and the behavior discontinuity marker, the state transition probability matrix is ​​used to dynamically simulate the discontinuous transition of vehicle behavior, generating multiple intention branch paths with corresponding probability weights. Finally, the intention branch paths are optimized based on distance metrics and corrected by behavior marker feedback. Step S3: Based on the generated intent branch path, correct non-standard behaviors, introduce avoidance simulation operation when judging the entry and exit intent, predict the interference amplification effect to isolate the incomplete behavior, and output the corrected intent vector. Step S4: Perform a record generation operation using the corrected intent vectors, adjust the isolation conditions of the intent vectors according to the prediction effect, and filter out the effective intent vectors from the intent vectors; calculate the proportion of the number of effective intent vectors in all corrected intent vectors, and compare it with the dynamic collaboration threshold to prevent collective misjudgment from causing a chain reaction of delays.

2. The parking lot management method for verifying the authenticity of vehicle entry and exit according to claim 1, characterized in that, Temporal analysis of location data is performed to calculate the difference in position coordinates between consecutive frames. This difference is then combined with the direction vector to determine jitter or reversal phenomena, forming a behavior discontinuity marker. Interference vectors are connected to the behavior discontinuity marker through grid temporal coupling calculation. The vector norm formula is used to mark the magnitude of position jitter or direction reversal at discontinuous points, avoiding inefficient identification caused by fragmented independent processing.

3. The parking lot management method for verifying the authenticity of vehicle entry and exit according to claim 1, characterized in that, Principal component analysis is used to reduce the dimensionality of the interference subspace. By calculating the covariance matrix of the interference vectors in the dataset, the eigenvalues ​​are obtained and sorted. The eigenvectors whose cumulative contribution rate of the eigenvalues ​​reaches a preset ratio are selected as principal components. The original interference vectors are projected onto these principal components to extract pure vectors. The elements of the covariance matrix are the calculation results of the variance and covariance between each dimension of the interference vector. The eigenvalues ​​are obtained by matrix diagonalization, and the projection operation is implemented by matrix multiplication.

4. The parking lot management method for verifying the authenticity of vehicle entry and exit according to claim 1, characterized in that, The process of generating intention branch paths and optimizing their weights and performing feedback correction includes: based on the extracted clean vector and behavior discontinuity markers, a Markov chain is used to simulate discontinuous behavior transitions to dynamically generate multiple intention branch paths; simultaneously, by comparing the Euclidean distance between the clean vector and each branch path, the gradient descent method is used to iteratively optimize the connection weights of the branch paths to minimize the distance error; and the magnitude of the behavior discontinuity markers is used as a feedback factor to correct the confidence of the corresponding state transitions in real time, thereby forming a complementary clean chain that integrates signal stripping, path generation, and feedback correction.

5. The parking lot management method for verifying the authenticity of vehicle entry and exit according to claim 1, characterized in that, Correcting non-standard behavior and predicting interference amplification effects includes: weighting and fusing the corresponding vectors of multiple intent branch paths using branch probabilities to form a single calibration intent vector; using Monte Carlo sampling, repeatedly and randomly adding neighboring vehicle interference variables following a preset distribution to the calibration intent vector to generate multiple simulated trajectories; in each sampling, a dynamic boundary threshold used to determine whether the simulated trajectory exceeds the limit is determined by the sum of a basic boundary threshold and the probability of the preceding interference amplification effect multiplied by a dynamic offset coefficient; the proportion of deviations exceeding the dynamic boundary threshold in all samplings is counted as the current probability of interference amplification effect; when the probability of interference amplification effect is higher than a preset boundary value, the corresponding intent vector is marked as an incomplete behavior and discarded.

6. The parking lot management method for verifying the authenticity of vehicle entry and exit according to claim 1, characterized in that, The intention vector is calibrated by using a branch probability weighted fusion operation. The probability value of each branch path is calculated as a weight, and the corresponding intention vectors in the path are weighted and summed. The weighted summation formula is that the corrected intention vector is equal to the sum of the products of each branch intention vector and its corresponding probability. Thus, multiple branches are fused to form a single calibrated intention vector. The intention vector weight of the low probability branch is multiplied by the reciprocal of the interference amplification effect probability for penalty adjustment. The reciprocal is calculated as 1 divided by the interference amplification effect probability. When the interference amplification effect probability is close to 1, the penalty magnitude increases, thereby achieving the correction of confusion for non-standard behaviors.

7. The parking lot management method for verifying the authenticity of vehicle entry and exit according to claim 1, characterized in that, The isolation condition for adjusting the intention vector based on the predicted effect refers to: based on the interference amplification effect probability calculated in step S3, the difference between it and the preset interference effect benchmark value is calculated, and then nonlinearly mapped through the hyperbolic tangent saturation function to obtain the adjustment coefficient; the adjustment coefficient is multiplied by the preset isolation threshold amplification coefficient, added to 1, and then multiplied by the original isolation threshold to obtain the adjusted isolation condition threshold; the adjusted isolation condition threshold is then used to perform geometric distance screening on the corrected intention vector to obtain the effective intention vector.

8. The parking lot management method for verifying the authenticity of vehicle entry and exit according to claim 1, characterized in that, The game theory equilibrium optimization in step S4 includes the following steps: A two-player game model is established between threshold adjustment strategy and interference environment state. The pure strategies of threshold adjustment strategy include relaxing the isolation threshold and tightening the isolation threshold, and the pure strategies of interference environment state include high interference scenario and low interference scenario. The revenue value corresponding to each strategy combination in the revenue matrix is ​​determined by statistical analysis of historical data or online learning. The revenue value is calculated by combining the recognition accuracy, false alarm rate and passage delay time according to preset weights. The Nash equilibrium point of the payoff matrix is ​​solved by iteratively eliminating inferior strategies to obtain the corresponding equilibrium solution. The equilibrium solution is mapped to the theoretical upper limit of the isolation condition threshold adjustment. During operation, if the threshold adjustment suggestion calculated based on the real-time interference amplification effect probability exceeds the theoretical upper limit, the theoretical upper limit is used as the actual adjustment result.

9. A parking lot management device for verifying the authenticity of vehicle entry and exit, comprising the method described in any one of claims 1-8, characterized in that, The method includes a processor, a memory, a communication interface, and one or more programs, said one or more programs being stored in the memory and configured to be executed by the processor, said programs including steps for performing the method as described in any one of claims 1-8.

10. A parking management system for verifying the authenticity of vehicle entry and exit, used to implement the method described in any one of claims 1-8, characterized in that, include: The trajectory extension module is used to capture surrounding interference vectors and behavioral discontinuity indicators through spatial grid division and temporal analysis mechanisms when reconstructing vehicle motion trajectories, forming an extended trajectory dataset. The interference vectors include overlapping contour point clouds, and the behavioral discontinuity indicators include position jitter or direction reversal. The intent purification module, based on an extended trajectory dataset, uses a vector interference stripping algorithm to separate the adjacent vehicle's ripple signal and integrates behavior probability calculation to dynamically generate intent branch paths; The risk prediction module corrects non-standard behaviors based on the generated intent branch path, introduces avoidance simulation operations when judging entry and exit intentions, predicts interference amplification effects to isolate incomplete behaviors, and outputs the corrected intent vector. The decision generation module uses the corrected intent vectors to perform record generation operations, adjusts the isolation conditions of the intent vectors according to the prediction effect, and filters out the effective intent vectors from the intent vectors; it calculates the proportion of the number of effective intent vectors in all corrected intent vectors and compares it with the dynamic collaboration threshold to prevent collective misjudgment from causing a chain reaction of delays.

Images