Intelligent parking unattended vehicle access management system and method based on edge computing

Abstract

The present application relates to the technical field of edge computing, in particular to a smart parking unattended vehicle access management system and method based on edge computing. It comprises: a vehicle detection and positioning unit acquires multi-modal data based on a multi-modal sensor array module, and realizes real-time positioning of the vehicle and real-time updating of the parking space state through Kalman filtering algorithm combined with QBCN information interference variable; a license plate recognition unit based on multi-spectral imaging technology and a lightweight CRNN model completes license plate feature extraction and recognition under abnormal lighting conditions; an edge computing decision unit combines a real-time rule engine and a spatio-temporal anomaly detection algorithm to optimize resource allocation strategies, realize low-latency decision-making and localized control; a payment authentication unit adopts a blockchain smart contract, dynamic rate calculation, multi-factor identity authentication and abnormal payment fuse protection; an edge cloud collaboration unit realizes model differential update, cross-domain task scheduling and energy efficiency optimization through federated learning and spatio-temporal data compression technology.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of edge computing, in particular to an intelligent parking unattended vehicle access management system and method based on edge computing. BACKGROUND

[0002] The intelligent parking unattended vehicle access management system based on edge computing is a solution that realizes real-time vehicle sensing, identity recognition and full-process automatic charging settlement by deploying edge computing nodes at the entrances and exits of parking lots and in parking space areas. The system uses the localized data processing capability of edge devices, combines license plate recognition algorithms, cloud account synchronization and Internet of Things gate control technologies, realizes automatic recognition of license plates, verification of permissions and lifting of gates when vehicles enter, and automatic deduction of fees based on parking time when vehicles leave by interacting with the payment platform through edge nodes, while encrypting and uploading core data to the cloud management platform. The core advantage is to reduce the load of cloud servers, reduce network delay, and ensure basic functions in the case of network interruption, thereby improving traffic efficiency, reducing labor costs and optimizing parking space utilization.

[0003] In the intelligent parking unattended vehicle access management system and method based on edge computing, complex environments cause sensor positioning drift, affecting real-time updating of parking space status; extreme light, obstruction or inclination angle cause a sharp drop in license plate recognition rate, causing billing disputes; high delay and resource competition caused by cloud dependence cannot meet real-time decision-making needs, especially during peak periods, response lag is prone to occur; there are risks of data tampering and unauthorized access in the payment link, and fixed rates cannot adapt to dynamic traffic demand; lack of coordination between multiple parking lot systems, redundant data transmission increases operation and maintenance costs, and lack of adaptive optimization mechanism in long-term evolution leads to low overall efficiency and limited scalability. Therefore, the intelligent parking unattended vehicle access management system and method based on edge computing are provided. SUMMARY

[0004] The present application aims to provide an intelligent parking unattended vehicle access management system and method based on edge computing to solve the problems of sensor positioning drift caused by complex environments, affecting real-time updating of parking space status; extreme light, obstruction or inclination angle causing a sharp drop in license plate recognition rate, causing billing disputes; high delay and resource competition caused by cloud dependence cannot meet real-time decision-making needs, especially during peak periods, response lag is prone to occur; there are risks of data tampering and unauthorized access in the payment link, and fixed rates cannot adapt to dynamic traffic demand; lack of coordination between multiple parking lot systems, redundant data transmission increases operation and maintenance costs, and lack of adaptive optimization mechanism in long-term evolution leads to low overall efficiency and limited scalability.

[0005] To achieve the above object, on the one hand, the application aims to provide an edge computing-based intelligent parking unattended vehicle access management system, comprising: a vehicle detection and positioning unit, which obtains multi-modal data based on a multi-modal sensor array module, and realizes real-time positioning of vehicles and real-time updating of parking space status through a Kalman filtering algorithm combined with QBCN information disturbance variables;

[0006] A license plate recognition unit, which is based on multi-spectral imaging technology and a lightweight CRNN model, and completes license plate feature extraction and recognition under abnormal lighting conditions, and supports multi-angle license plate correction through vehicle pose estimation technology;

[0007] An edge computing decision unit, which combines a real-time rule engine and a spatio-temporal anomaly detection algorithm to dynamically optimize resource allocation strategies, realize low-latency decision-making and localized control;

[0008] A payment authentication unit, which uses a blockchain smart contract to support dynamic rate calculation, multi-factor identity authentication and abnormal payment fuse protection;

[0009] An edge cloud collaboration unit, which realizes model differential update, cross-domain task scheduling and energy efficiency optimization through federated learning and spatio-temporal data compression technology.

[0010] As a further improvement of the technical solution, the multi-modal sensor array module comprises a camera module, an infrared sensor module and a radar sensor module;

[0011] The camera module is used for high-resolution image acquisition in a visible light environment, and supports color feature extraction and preliminary positioning;

[0012] The infrared sensor module detects the existence of vehicles and enhances the outline under low-visibility conditions by perceiving the temperature distribution of vehicles through thermal radiation;

[0013] The radar sensor module accurately obtains the real-time speed, distance and azimuth angle of vehicles based on millimeter wave Doppler effect and time-of-flight measurement.

[0014] As a further improvement of the technical solution, the vehicle detection and positioning unit further comprises a target positioning algorithm module, and the specific process of the target positioning algorithm module to realize real-time positioning of vehicles and real-time updating of parking space status through a Kalman filtering algorithm combined with QBCN information disturbance variables is as follows:

[0015] The multi-modal data obtained based on the multi-modal sensor array module is fused through a weighted fusion algorithm to obtain fused multi-modal data , and normalized to obtain normalized multi-modal data ;

[0016] According to the normalized multi-modal data Through the Kalman filtering algorithm and combined with the QBCN information disturbance variable Get the optimized parking space state estimation value .

[0017] As a further improvement of the technical solution, the license plate recognition unit includes a multispectral imaging module, a feature extraction module, and a multi-license plate correction module.

[0018] The multispectral imaging module fuses multispectral image data under abnormal lighting conditions based on multispectral imaging technology .

[0019] The feature extraction module extracts license plate character sequence features based on a lightweight CRNN model.

[0020] The multi-license plate correction module implements multi-angle license plate correction based on vehicle pose estimation technology.

[0021] As a further improvement of the technical solution, the multi-license plate correction module supports multi-angle license plate correction through vehicle pose estimation technology, and the specific steps are as follows:

[0022] According to the normalized multi-modal data obtained by the target positioning algorithm module Compensate for license plate tilt caused by vehicle turning through vehicle pose estimation technology.

[0023] Fuse multispectral image data Through the projection matrix Project to the orthographic plane to obtain the orthographic license plate image .

[0024] As a further improvement of the technical solution, the edge computing decision unit includes a real-time rule engine module, a spatio-temporal anomaly detection module, and a resource dynamic scheduling module.

[0025] The real-time rule engine module executes immediate control logic based on parking space state and vehicle pose.

[0026] The spatio-temporal anomaly detection module is used to detect abnormal patterns of vehicle behavior and sensor data.

[0027] The resource dynamic scheduling module is used to dynamically adjust system resource allocation according to the comprehensive anomaly score And the rule function Output the control action vector Through the bisection algorithm.

[0028] As a further improvement of the technical solution, the specific steps of the spatio-temporal anomaly detection module detecting abnormal patterns of vehicle behavior and sensor data are:

[0029] According to the normalized multi-modal data And the parking space state estimation value Trajectory deviation detection and sensor data consistency detection are performed through an anomaly detection algorithm;

[0030] The comprehensive anomaly score is calculated according to the results of trajectory deviation detection and sensor data consistency detection .

[0031] As a further improvement of the technical solution, the payment authentication unit includes a blockchain smart contract module, a multi-factor identity authentication module, and a payment fuse protection module;

[0032] The blockchain smart contract module is used to automatically execute parking fee settlement and transaction record chaining;

[0033] The multi-factor identity authentication module is used to combine license plate recognition, facial features, and bioelectric signals to authenticate user identity;

[0034] The payment fuse protection module generates a comprehensive risk score through weighted multi-dimensional abnormal indicators , and automatically triggers the fuse mechanism when abnormal payment behavior occurs.

[0035] As a further improvement of the technical solution, the edge cloud collaboration unit includes a model differential federation module, a spatio-temporal data compression module, and a collaborative task scheduling module;

[0036] The model differential federation module is used for differential update between local models and cloud aggregated models;

[0037] The spatio-temporal data compression module is used to compress the normalized multi-modal data using a compression algorithm with a Transformer structure;

[0038] The collaborative task scheduling module dynamically allocates tasks to optimize overall system efficiency through a total cost minimization algorithm based on local and cloud resource conditions.

[0039] On the other hand, the present application provides a smart parking unattended vehicle access management method based on edge computing, based on the above-mentioned smart parking unattended vehicle access management system based on edge computing, including the following steps:

[0040] S10.1, real-time acquisition of environmental data by a multi-modal sensor array, fusion of QBCN interference compensation algorithm through Kalman filtering, output of vehicle coordinates and dynamic update of parking space occupation state;

[0041] S10.2, capture of multi-dimensional features of license plates based on multi-spectral imaging, cross-illumination scene recognition using a lightweight CRNN model, geometric correction and analysis of tilted and occluded license plates by combining vehicle pose estimation;

[0042] S10.3, integration of real-time rule engine and spatio-temporal anomaly detection model, allocation of computing resources through dynamic game optimization algorithm, realization of millisecond-level localized decision and system anomaly self-healing control;

[0043] S10.4, automatic execution of dynamic rate calculation and multi-factor identity verification relying on a blockchain smart contract, triggering of a hierarchical fuse mechanism through transaction behavior analysis for the purpose of ensuring payment security and intercepting abnormal transactions;

[0044] S10.5, use of federated learning to aggregate multi-node feature differences, combined with spatio-temporal compression coding technology to realize efficient model updating and task scheduling.

[0045] Compared with the prior art, the beneficial effects of the present application are:

[0046] 1. In the edge computing-based intelligent parking unattended vehicle access management system and method, the vehicle position is sensed in real time by a multi-modal sensor array, combined with Kalman filtering and QBCN anti-interference algorithm, the vehicle trajectory is dynamically tracked and the parking space occupation state is accurately updated, providing bottom layer spatial perception data for the system. Based on multi-spectral imaging and a lightweight CRNN model, license plate feature extraction and correction recognition are completed under complex illumination / angle conditions, and standardized vehicle identity information is output to support billing and security authentication processes.

[0047] 2. In the edge computing-based intelligent parking unattended vehicle access management system and method, a real-time rule engine and a spatio-temporal anomaly detection algorithm are integrated, the edge node resource allocation is dynamically optimized, millisecond-level localized decision is realized, and low-latency response of the system is ensured. Relying on a blockchain smart contract, dynamic rate calculation and multi-factor identity verification are automatically executed, a fuse protection mechanism is triggered through transaction behavior analysis to ensure payment security and intercept abnormal transactions. Through federated learning to aggregate multi-node feature differences, combined with spatio-temporal data compression technology, model differential updating and cross-domain task scheduling are realized to complete system-level energy optimization and long-term performance evolution. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 The overall flowchart of the present application is shown in the figure;

[0049] The meanings of the various labels in the figure are as follows:

[0050] 1, vehicle detection positioning unit; 11, multi-modal sensor array module; 111, camera module; 112, infrared sensor module; 113, radar sensor module; 12, target positioning algorithm module; 2, license plate recognition unit; 21, multi-spectral imaging module; 22, feature extraction module; 23, multi-license plate correction module; 3, edge computing decision unit; 31, real-time rule engine module; 32, spatio-temporal anomaly detection module; 33, resource dynamic scheduling module; 4, payment authentication unit; 41, blockchain smart contract module; 42, multi-factor identity authentication module; 43, payment fuse protection module; 5, edge cloud collaboration unit; 51, model differential federation module; 52, spatio-temporal data compression module; 53, collaborative task scheduling module. DETAILED DESCRIPTION

[0051] The technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0052] Embodiment 1: Please refer to Figure 1 As shown in the figure, an edge computing-based intelligent parking unattended vehicle access management system is provided, which comprises a vehicle detection positioning unit 1. The vehicle detection positioning unit 1 acquires multi-modal data based on a multi-modal sensor array module 11, and realizes real-time positioning of vehicles through Kalman filtering algorithm, dynamically tracks the motion trajectory of multiple targets and updates the parking space status in real time.

[0053] In this example, the multi-modal sensor array module 11 comprises a camera module 111, an infrared sensor module 112 and a radar sensor module 113.

[0054] The camera module 111 is used for high-resolution image acquisition in a visible light environment, supporting vehicle appearance recognition, color feature extraction and preliminary positioning.

[0055] The infrared sensor module 112 perceives the temperature distribution of vehicles through thermal radiation, realizes vehicle existence detection and contour enhancement under low visibility conditions.

[0056] The radar sensor module 113 accurately acquires the real-time speed, distance and azimuth angle of vehicles based on millimeter wave Doppler effect and time-of-flight measurement.

[0057] In the present example, the vehicle detection and positioning unit 1 further comprises a target positioning algorithm module 12, which implements real-time positioning of vehicles by means of a Kalman filtering algorithm, dynamically tracks the motion trajectories of multiple targets and updates the state of the parking space in real time. The specific process is as follows:

[0058] The multi-modal data obtained based on the multi-modal sensor array module 11 is fused by a weighted fusion algorithm to obtain fused multi-modal data , and normalized to obtain normalized multi-modal data ;

[0059] Among them, the multi-modal data includes image data , infrared data and vehicle dynamic data ;

[0060] Specifically, the fused multi-modal data is:

[0061]

[0062] In the formula, t represents the time; wi represents the weight coefficient of the image data; wd represents the weight coefficient of the vehicle dynamic data; wi represents the weight coefficient of the infrared data; and ;

[0063] Among them, the image data is:

[0064] ;

[0065] In the formula, vi represents the visual positioning coordinates; vi represents the visual velocity estimation based on optical flow; ci represents the color feature;

[0066] The vehicle dynamic data is:

[0067] ;

[0068] In the formula, ri represents the position in the radar global coordinate system; vi represents the Doppler velocity; ai represents the azimuth angle;

[0069] The infrared data is:

[0070] ;

[0071] In the formula, Indicates the coordinates of the thermal imaging location; This indicates the average temperature in the vehicle area. Indicates the existence confidence level. ;

[0072] in, The relationship with the existence confidence level is as follows:

[0073] ;

[0074] In the formula, Indicates the basic weight value; This indicates the maximum value of the specified weight;

[0075] For multimodal data Normalization is performed:

[0076] ;

[0077] In the formula, This represents multimodal data after normalization. This represents the minimum value in the dataset; This represents the maximum value in the dataset;

[0078] Based on the normalized multimodal data The Kalman filter algorithm is used in conjunction with QBCN information interference variables. Obtain the optimized parking space status estimate .

[0079] Specifically, based on the normalized multimodal data Parking space status estimates are obtained using the Kalman filter algorithm. The steps are as follows:

[0080] Define state vector :

[0081] ;

[0082] In the formula, Indicates the position in the global coordinate system; Indicates global speed; Indicates the global heading angle;

[0083] Then, based on the normalized multimodal data Define observation vector for:

[0084] ;

[0085] In the formula, Indicates the position in the observation coordinate system; Indicates instantaneous velocity; Indicates the observed heading angle;

[0086] Among them, the observation vector The variables in the table are:

[0087] ;

[0088] ;

[0089] ;

[0090] ;

[0091] In the formula, Indicates infrared velocity;

[0092] in, for:

[0093] ;

[0094] In the formula, This represents the rate of change of the source region temperature over time. Represents the spatial gradient of temperature; A unit vector representing the direction of the temperature gradient;

[0095] Parking space status estimates are obtained using the Kalman filter algorithm. ;

[0096] but for:

[0097] ;

[0098] In the formula, Indicates at time The predicted value of the prior state; Indicates Kalman gain; Represents the observation matrix;

[0099] in, for:

[0100] ;

[0101] In the formula, Represents the state transition matrix; Indicates at time posterior state predictions;

[0102] Wherein, the state transition matrix for:

[0103] ;

[0104] In the formula, Indicates time The global heading angle; Indicates time Global speed;

[0105] Kalman gain for:

[0106] ;

[0107] In the formula, Indicates covariance prediction; Indicates the observation noise covariance; Indicates the transpose operation;

[0108] Among them, covariance prediction for:

[0109] ;

[0110] In the formula, Represents the process noise covariance matrix; Indicates time The posterior estimation error covariance;

[0111] Among them, the process noise covariance matrix for:

[0112] ;

[0113] In the formula, The positional noise variance of the horizontal axis; This represents the positional noise variance of the ordinate. Indicates the variance of velocity noise; Indicates the variance of the direction angle noise;

[0114] Among them, the observation matrix for:

[0115] ;

[0116] Considering that information interference variables can affect the parking space status estimate To improve the accuracy, QBCN information interference variables are introduced to optimize the parking space status estimate. The result;

[0117] The optimized parking space state estimate for:

[0118] ;

[0119] In the formula, Indicates the interference variable of QBCN information;

[0120] Among them, QBCN information interference variables for:

[0121] ;

[0122] In the formula, This represents quantum tunneling transient noise; Indicates bioelectrical rhythm noise; Indicates the amplitude of quantum tunneling noise; Indicates the time width of quantum noise; Indicates the frequency of bioelectrical rhythms; This indicates the central time point at which quantum tunneling interference occurs; This represents the initial phase angle of a sine wave;

[0123] Specifically, in radar sensor module 113, quantum tunneling can cause instantaneous current pulses, leading to abrupt changes in position / velocity observations.

[0124] Bioelectric interference occurs when the driver's / passenger's bioelectrical signals (such as electrocardiogram and electromyography) are coupled to the vehicle's CAN bus through the onboard biosensor system, periodically affecting sensor data.

[0125] Furthermore, by comprehensively utilizing multimodal sensor data fusion, Kalman filtering algorithms, and considering information interference variables (such as quantum tunneling transient noise and bioelectrical rhythm noise) to optimize parking space status estimates, accurate tracking and real-time updates of the dynamic characteristics of vehicles within the parking lot, including position, speed, and direction, are achieved. This mechanism not only improves the accuracy of parking space occupancy assessment but also effectively addresses errors caused by sensor noise and other environmental factors, ensuring efficient and reliable management of parking resources, thereby enhancing user experience and maximizing parking lot operational efficiency. In addition, through the application of edge computing technology, the system can rapidly process large amounts of sensor data locally, reducing latency and further enhancing system response speed and stability.

[0126] The edge computing-based smart parking unattended vehicle access management system also includes a license plate recognition unit 2. The license plate recognition unit 2 is based on multispectral imaging technology and a lightweight CRNN model. It completes license plate feature extraction and recognition under abnormal lighting conditions and supports multi-angle license plate correction through vehicle pose estimation technology.

[0127] In this example, the license plate recognition unit 2 includes a multispectral imaging module 21, a feature extraction module 22, and a multi-license plate correction module 23;

[0128] The multispectral imaging module 21 fuses multispectral image data under abnormal lighting conditions based on multispectral imaging technology. ;

[0129] Specifically, based on multispectral imaging technology, multispectral image data is fused under abnormal lighting conditions. The process is as follows:

[0130] In cases of abnormal lighting, artificial strong light irradiation can affect image sampling and feature extraction. Therefore, an abnormal lighting coefficient is introduced. The results of each module are optimized in the multispectral imaging module 21, the feature extraction module 22, and the multi-license plate correction module 23.

[0131] Visible light images are acquired using camera module 111. Near-infrared images are acquired by infrared sensor module 112. Through multispectral imaging technology combined with illumination anomaly coefficient Multispectral images are obtained by fusion ;

[0132] Multispectral images for:

[0133] ;

[0134] In the formula, This represents the visible light weighting coefficient; Indicates the near-infrared weighting coefficient; This indicates a dynamically compressed visible light image;

[0135] Among them, dynamically compressed visible light images for:

[0136] ;

[0137] In the formula, Indicates the compressive strength coefficient; Indicates compensation bias;

[0138] Among them, the light anomaly coefficient for:

[0139] ;

[0140] In the formula, Representing visible light images Near-infrared images Histogram differences;

[0141] when At that time, it was either low light or normal light.

[0142] when At that time, it was either strong light or extreme light.

[0143] The feature extraction module 22 extracts license plate character sequence features based on a lightweight CRNN model;

[0144] Specifically, the process of building a lightweight CRNN model is as follows:

[0145] Lightweight CRNN models are divided into:

[0146] Convolutional layers: convert multispectral images Vehicle color characteristics splicing output feature map ;

[0147] In the formula, Indicates a splicing operation; Represents the set of real numbers; Indicates altitude; Indicates width; Indicates the number of channels;

[0148] Bidirectional LSTM layer: Output temporal features ;

[0149] In the formula, This represents a bidirectional long short-term memory network;

[0150] CTC Decoding Layer: Output Character Sequence Features ;

[0151] The characteristics of the character sequence are:

[0152] ;

[0153] In the formula, This indicates the connection timing classification and decoding process; This represents the weight matrix of the fully connected layer; This means converting the input into a probability distribution; This represents the bias vector of the fully connected layer;

[0154] Among them, the weight matrix of the fully connected layer for:

[0155] ;

[0156] In the formula, Indicates the feature dimension; Indicates the size of the character set;

[0157] The multi-license plate correction module 23 achieves multi-angle license plate correction based on vehicle pose estimation technology.

[0158] In this example, the specific steps of the multi-license plate correction module 23 supporting multi-angle license plate correction through vehicle pose estimation technology are as follows:

[0159] Based on the multimodal data obtained from the target localization algorithm module 12 Vehicle pose estimation technology is used to compensate for license plate tilt caused by vehicle steering.

[0160] Specifically, the process of compensating for license plate tilt caused by vehicle steering using vehicle pose estimation technology is as follows:

[0161] Based on the state vector and the original license plate corner coordinates obtained through camera module 111 Transform the corner points from the image coordinate system to the vehicle body coordinate system, and output the corrected corner point coordinates. ;

[0162] The conversion formula is:

[0163] ;

[0164] In the formula, Indicates a position index variable;

[0165] Multispectral image data Through projection matrix Projecting the image onto the frontal plane yields a frontalized license plate image. ;

[0166] Specifically, multispectral image data Through projection matrix The process of projecting onto the viewing plane is as follows:

[0167] Based on multispectral image data and corrected corner coordinates Solving the projection matrix using the least squares method ,satisfy:

[0168] ;

[0169] In the formula, This represents the coordinates of the corner points of a standard license plate.

[0170] The corrected image is generated using a bilinear interpolation algorithm;

[0171] Specifically, for each pixel of the corrected image Calculate its coordinates in the original image. ;

[0172] ;

[0173] In the formula, The normalization factor represents the homogeneous coordinates;

[0174] Output of the frontalized license plate image for:

[0175] ;

[0176] In the formula, Represents the pixel index variable;

[0177] Furthermore, the multispectral imaging module 21 fuses visible and near-infrared images and dynamically adjusts image quality using an illumination anomaly coefficient, effectively suppressing strong light interference. Then, the feature extraction module 22 extracts license plate character sequence features based on a lightweight CRNN model, achieving high-precision character recognition. Finally, the multi-license plate correction module 23 uses vehicle pose estimation technology to geometrically correct license plates that are tilted or have changing angles, generating a normalized license plate image through coordinate transformation and bilinear interpolation. This ensures clear and standard license plate images are obtained even when the vehicle is turning or tilting. The entire process significantly enhances the system's license plate recognition capabilities in all-weather, multi-angle, and complex environments, providing reliable technical support for automatic vehicle identity verification and access control in unattended scenarios.

[0178] The intelligent parking unattended vehicle access management system based on edge computing also includes an edge computing decision unit 3. The edge computing decision unit 3 combines a real-time rule engine and a spatiotemporal anomaly detection algorithm to dynamically optimize resource allocation strategies and achieve low-latency decision-making and localized control.

[0179] In this example, the edge computing decision unit 3 includes a real-time rule engine module 31, a spatiotemporal anomaly detection module 32, and a resource dynamic scheduling module 33;

[0180] The real-time rule engine module 31 executes real-time control logic based on the parking space status and vehicle position.

[0181] Specifically, based on the estimated value of the parking space status Vehicle position information Features of license plate character sequences Constructing real-time rule signals ;

[0182] but for:

[0183] ;

[0184] In the formula, Represents a rule function;

[0185] The logic for judging regular signals is as follows:

[0186] ;

[0187] In the formula, Represents a logical region; This indicates that the requirements are met and the vehicle is currently parking. This indicates that the requirements are not met and parking should be stopped.

[0188] The spatiotemporal anomaly detection module 32 is used to detect abnormal patterns in vehicle behavior and sensor data;

[0189] In this example, the specific steps of the spatiotemporal anomaly detection module 32 in detecting abnormal patterns in vehicle behavior and sensor data are as follows:

[0190] Based on the normalized multimodal data Parking space status estimate Trajectory deviation detection and sensor data consistency detection are performed using anomaly detection algorithms.

[0191] Specifically, the process of using anomaly detection algorithms to perform trajectory deviation detection and sensor data consistency detection is as follows:

[0192] Trajectory deviation detection is as follows:

[0193] ;

[0194] In the formula, Represents the residual between the prediction and the observation; This represents the predicted vehicle trajectory value;

[0195] in, for:

[0196] ;

[0197] In the formula, Indicates time Estimated parking space status;

[0198] when The trajectory was determined to be abnormal. Indicates the standard residual;

[0199] Sensor data consistency detection is as follows:

[0200] ;

[0201] In the formula, This indicates the definition of the consistency index value between sensors; Represents standard sensor data;

[0202] Among them, standard sensor data Obtained through expert experience;

[0203] when The sensor was determined to be faulty. This indicates the consistency index value among standard sensors;

[0204] A comprehensive anomaly score is calculated based on the results of trajectory deviation detection and sensor data consistency detection. .

[0205] Specifically, for:

[0206] ;

[0207] In the formula, express Weighting coefficients; express Weighting coefficients;

[0208] when When this occurs, an abnormal alarm is triggered;

[0209] The resource dynamic scheduling module 33 is used to perform comprehensive anomaly scoring based on the comprehensive anomaly score. and rule functions The control action vector is output using the least binary search algorithm. Dynamically adjust system resource allocation.

[0210] Specifically, control action vectors for:

[0211]

[0212] In the formula, This represents the mapping from actions to the task space; Indicates the initial control action vector; Represents the regularization parameter; The penalty coefficient for the control action is used to adjust the relative weight of the control term in the optimization function. Its value can be dynamically set according to the system state, scheduling strategy or resource availability; u represents a control action vector to be optimized, representing a specific edge control action, such as: activation and angle adjustment of camera / infrared devices, gate lifting control, communication resource (such as bandwidth / frequency) allocation strategy, and priority setting of data processing module.

[0213] Furthermore, by constructing real-time rule signals to determine whether vehicles comply with parking rules (such as parking space occupancy status, vehicle position and license plate characteristics), and combining trajectory deviation detection and sensor consistency analysis to identify potential abnormal behaviors or data errors, the robustness and security of the system are improved. Finally, the resource dynamic scheduling module generates the optimal control action vector based on the rule judgment results and anomaly scores using optimization algorithms, realizing intelligent scheduling and priority management of system resources (such as cameras, gates, communication bandwidth, etc.), ensuring localized decision-making and control under low latency, and significantly improving the operating efficiency and automation level of the parking lot.

[0214] The edge computing-based smart parking unmanned vehicle access management system also includes a payment authentication unit 4. The payment authentication unit 4 adopts a blockchain smart contract and supports dynamic rate calculation, multi-factor authentication, and circuit breaker protection for abnormal payments.

[0215] In this example, the payment authentication unit 4 includes a blockchain smart contract module 41, a multi-factor authentication module 42, and a payment circuit breaker protection module 43.

[0216] The blockchain smart contract module 41 is used to automatically execute parking fee settlement and transaction record uploading on the blockchain;

[0217] Specifically, parking fee settlement for:

[0218] ;

[0219] In the formula, Indicates the time of departure; Indicates the entry time; Indicates the parking space level; Indicates the rate volatility factor over a specific time period;

[0220] The multi-factor authentication module 42 is used to combine license plate recognition, facial features and bioelectric signals to perform reliable authentication of user identity.

[0221] Specifically, the license plate recognition unit extracts the character sequence of the vehicle's license plate. The system compares the license plate information with the pre-stored user-bound license plate information to determine if the license plate is on the authorized whitelist. It also captures the driver's or user's facial image through a camera and compares it with the facial feature database of registered users in the system. If the authentication is successful, the payment process is allowed; otherwise, the transaction is rejected and a manual review is prompted.

[0222] The payment circuit breaker protection module 43 generates a comprehensive risk score by weighting multi-dimensional anomaly indicators. And automatically trigger the circuit breaker mechanism when abnormal payment behavior occurs.

[0223] Specifically, comprehensive risk score for:

[0224] ;

[0225] In the formula, Indicates the first The actual value of the abnormal indicator; Indicates the first Thresholds for each indicator; express Weighting coefficients; Index variables representing abnormal indicators; This indicates the total amount of abnormal indicators;

[0226] The automatic circuit breaker mechanism is as follows:

[0227] ;

[0228] In the formula, Indicates a circuit breaker operation;

[0229] when At that time, payment is restricted; Payment is allowed at that time;

[0230] Specifically, when one or more risk indicators are detected to exceed the preset security threshold, the system automatically assesses whether there is fraud or illegal transaction behavior; once a high-risk behavior is determined, the payment circuit breaker mechanism is immediately triggered, suspending the current transaction process and sending an alarm message to the management backend, while recording complete transaction context data for subsequent auditing.

[0231] Furthermore, in the edge computing-based smart parking unmanned vehicle access management system, a blockchain smart contract module, a multi-factor authentication module, and a payment circuit breaker protection module jointly construct a secure, reliable, and automated payment and identity management system. Specifically, the blockchain smart contract module automatically executes parking fee settlements and records transaction data on the blockchain, ensuring transparent and tamper-proof billing and enhancing the system's credibility. The multi-factor authentication module integrates license plate recognition, facial features, and bioelectrical signals to achieve highly reliable user identity verification, preventing impersonation and unauthorized access. The payment circuit breaker protection module automatically triggers a circuit breaker mechanism when abnormal payment behavior is detected, promptly blocking risky transactions and ensuring fund security. The synergistic effect of these three modules effectively supports the security, compliance, and efficient operation of the payment process in unmanned scenarios.

[0232] The edge computing-based intelligent parking unmanned vehicle access management system also includes an edge-cloud collaboration unit 5. This edge-cloud collaboration unit 5 uses federated learning and spatiotemporal data compression technology to achieve model differential updates, cross-domain task scheduling, and energy efficiency optimization.

[0233] In this example, the edge-cloud collaboration unit 5 includes a model differential federation module 51, a spatiotemporal data compression module 52, and a collaborative task scheduling module 53;

[0234] The model differential federation module 51 is used to perform differential updates between the local model and the cloud-aggregated model.

[0235] Specifically, after training the model on the local device, the original data is not uploaded. Instead, the differences between the local model parameters and the global model in the cloud are encrypted and compressed, and key gradient directions are selected for sparsification before being uploaded to the cloud for aggregation and update.

[0236] Calculate the difference ;

[0237] ;

[0238] In the formula, Indicates the first Local model parameters for each edge node; Represents the edge node index variable; Indicates the current global model parameters;

[0239] Select the vector with the largest absolute value. Each element is the part to be uploaded;

[0240] The difference vector after sparsification is obtained by Top-K selection. ;

[0241] ;

[0242] After receiving the model difference information uploaded by all edge nodes, the cloud uses a weighted average algorithm to generate a new global model;

[0243] Then the new global model parameters for:

[0244] ;

[0245] In the formula, This represents the total number of edge nodes; Indicates the first The weight coefficients of each edge node;

[0246] The spatiotemporal data compression module 52 is used to apply a compression algorithm based on the Transformer structure to the normalized multimodal data. Compress;

[0247] A compression algorithm based on the Transformer architecture is used to process the normalized multimodal data. The numerical expression for compression is:

[0248] ;

[0249] In the formula, This represents compressed multimodal data; This describes the entire operation flow of the Transformer encoder;

[0250] The entire operation process of the Transformer encoder includes:

[0251] Add positional encoding: Add positional information to the input sequence;

[0252] Multi-head self-attention mechanism: capturing long-distance dependencies in input data;

[0253] Feedforward neural networks: further process the feature representations at each time step;

[0254] Residual connections and layer normalization: stabilize the training process and accelerate convergence.

[0255] Specifically, the high-dimensional perception data from the vehicle detection and positioning unit and the license plate recognition unit is compressed and encoded to reduce the communication bandwidth pressure between edge devices and the cloud. This module uses a Transformer-based compression algorithm to perform low-dimensional embedding representation on the input spatiotemporal data stream and output a compressed data stream. At the same time, it ensures that the reconstruction error between the decompressed data and the original data is below a preset threshold, thereby significantly reducing transmission overhead while ensuring the integrity of key information.

[0256] The collaborative task scheduling module 53 dynamically allocates tasks based on local and cloud resource conditions and optimizes overall system efficiency through a total cost minimization algorithm.

[0257] The numerical expression for optimizing overall system efficiency by dynamically allocating tasks using a total cost minimization algorithm is:

[0258] ;

[0259] In the formula, This represents the minimum total cost; Indicates task The computational complexity; Indicates the priority of the task; Indicates the cost of cloud communication; Represent decision variables; This represents the task index variable;

[0260] When the task is executed locally, then When the task is executed in the cloud, then .

[0261] Specifically, the computational complexity and priority of each task in the current task queue are evaluated in real time. Combined with local energy consumption and cloud communication costs, it is determined whether each task is executed locally or uploaded to the cloud for processing. The final task scheduling scheme maximizes resource utilization and optimizes energy efficiency while ensuring response latency, thereby improving the system's flexibility and scalability.

[0262] Furthermore, in the edge computing-based intelligent parking unmanned vehicle access management system, the edge-cloud collaboration unit integrates a model differential federation module, a spatiotemporal data compression module, and a collaborative task scheduling module to achieve efficient collaboration and resource optimization between the local and cloud environments. Specifically, the model differential federation module supports differential updates between the local model and the cloud-aggregated model, improving model accuracy while ensuring data privacy and system learning efficiency; the spatiotemporal data compression module compresses high-dimensional sensing data (such as images, radar, pose, etc.), reducing communication bandwidth pressure and improving transmission efficiency; and the collaborative task scheduling module intelligently allocates computing tasks based on the resource load of local edge nodes and the cloud, balancing workloads and improving the overall system's response speed and operating efficiency. This collaborative mechanism effectively supports intelligent and distributed vehicle management and decision-making capabilities in large-scale parking lot environments.

[0263] Example 2:

[0264] The difference between Embodiment 2 and Embodiment 1 is that this embodiment introduces the static mechanical performance acquisition and analysis method used in the intelligent parking unattended vehicle access management system based on edge computing.

[0265] A smart parking unattended vehicle access management method based on edge computing, based on the aforementioned smart parking unattended vehicle access management system based on edge computing, includes the following steps:

[0266] S10.1. Real-time environmental data is collected through a multimodal sensor array, processed by Kalman filtering and QBCN interference compensation algorithm, and the vehicle coordinates are output and the parking space occupancy status is dynamically updated.

[0267] S10.2. Based on multispectral imaging to capture multi-dimensional features of license plates, a lightweight CRNN model is used for cross-illumination scene recognition, and combined with vehicle pose estimation to achieve geometric correction and analysis of tilted and occluded license plates.

[0268] S10.3 integrates a real-time rule engine and a spatiotemporal anomaly detection model, and allocates computing resources through a dynamic game optimization algorithm to achieve millisecond-level localized decision-making and system anomaly self-healing control.

[0269] S10.4. Relying on blockchain smart contracts to automatically execute dynamic fee calculation and multi-factor authentication, a tiered circuit breaker mechanism is triggered through transaction behavior analysis to ensure payment security and intercept abnormal transactions.

[0270] S10.5. Federated learning is used to aggregate the feature differences of multiple nodes, and spatiotemporal compression coding technology is combined to achieve efficient model updates and task scheduling.

[0271] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A smart parking unattended vehicle access management system based on edge computing, characterized in that, include: The vehicle detection and positioning unit (1) acquires multimodal data based on the multimodal sensor array module (11), and achieves real-time vehicle positioning and real-time updating of parking space status through Kalman filtering algorithm and QBCN information interference variables. The license plate recognition unit (2) is based on multispectral imaging technology and a lightweight CRNN model. It completes the extraction and recognition of license plate features under abnormal lighting conditions and supports multi-angle license plate correction through vehicle pose estimation technology. Edge computing decision unit (3), which combines real-time rule engine and spatiotemporal anomaly detection algorithm to dynamically optimize resource allocation strategy and realize low-latency decision-making and localized control; Payment authentication unit (4) adopts blockchain smart contract and supports dynamic rate calculation, multi-factor authentication and abnormal payment circuit breaker protection; Edge-cloud collaboration unit (5), which realizes model differential update, cross-domain task scheduling and energy efficiency optimization through federated learning and spatiotemporal data compression technology; The edge computing decision unit (3) includes a real-time rule engine module (31), a spatiotemporal anomaly detection module (32), and a resource dynamic scheduling module (33). The real-time rule engine module (31) executes real-time control logic based on the parking space status and vehicle position; The spatiotemporal anomaly detection module (32) is used to detect abnormal patterns in vehicle behavior and sensor data; The resource dynamic scheduling module (33) is used to perform comprehensive anomaly scoring. and rule functions The control action vector is output using the least binary search algorithm. Dynamically adjust system resource allocation; The specific steps of the spatiotemporal anomaly detection module (32) in detecting abnormal patterns in vehicle behavior and sensor data are as follows: Based on the normalized multimodal data Parking space status estimate Trajectory deviation detection and sensor data consistency detection are performed using anomaly detection algorithms. A comprehensive anomaly score is calculated based on the results of trajectory deviation detection and sensor data consistency detection. ; The edge-cloud collaboration unit (5) includes a model differential federation module (51), a spatiotemporal data compression module (52), and a collaboration task scheduling module (53). The model differential federation module (51) is used to perform differential updates between the local model and the cloud-aggregated model. The spatiotemporal data compression module (52) is used to apply a compression algorithm based on the Transformer structure to the normalized multimodal data. Compress; The collaborative task scheduling module (53) dynamically allocates tasks and optimizes the overall system efficiency based on the local and cloud resource status and through a total cost minimization algorithm. The numerical expression for optimizing overall system efficiency by dynamically allocating tasks using a total cost minimization algorithm is: ； In the formula, This represents the minimum total cost; Indicates task The computational complexity; Indicates the priority of the task; Indicates the cost of cloud communication; Represent decision variables; This represents the task index variable.

2. The intelligent parking unattended vehicle access management system based on edge computing as described in claim 1, characterized in that: The multimodal sensor array module (11) includes a camera module (111), an infrared sensor module (112), and a radar sensor module (113). The camera module (111) is used for high-resolution image acquisition in visible light environments and supports color feature extraction and preliminary positioning. The infrared sensor module (112) senses the vehicle temperature distribution through thermal radiation, thereby realizing vehicle presence detection and contour enhancement under low visibility conditions. The radar sensor module (113) accurately acquires the vehicle's real-time speed, distance, and azimuth angle based on the millimeter-wave Doppler effect and time-of-flight measurement.

3. The intelligent parking unattended vehicle access management system based on edge computing as described in claim 2, characterized in that: The vehicle detection and positioning unit (1) also includes a target positioning algorithm module (12). The specific process of the target positioning algorithm module (12) realizing real-time vehicle positioning and real-time updating of parking space status through Kalman filtering algorithm and QBCN information interference variables is as follows: Multimodal data acquired by the multimodal sensor array module (11) is fused using a weighted fusion algorithm to obtain fused multimodal data. The data was then normalized to obtain the normalized multimodal data. ; Based on the normalized multimodal data The Kalman filter algorithm is used in conjunction with QBCN information interference variables. Obtain the optimized parking space status estimate .

4. The intelligent parking unattended vehicle access management system based on edge computing according to claim 3, characterized in that: The license plate recognition unit (2) includes a multispectral imaging module (21), a feature extraction module (22), and a multi-license plate correction module (23). The multispectral imaging module (21) is based on multispectral imaging technology and combined with illumination anomaly coefficient. Fusion of multispectral image data under abnormal lighting conditions ; The feature extraction module (22) extracts license plate character sequence features based on a lightweight CRNN model; The multi-license plate correction module (23) realizes multi-angle license plate correction based on vehicle pose estimation technology.

5. The intelligent parking unattended vehicle access management system based on edge computing according to claim 4, characterized in that, The specific steps of the multi-license plate correction module (23) supporting multi-angle license plate correction through vehicle pose estimation technology are as follows: Based on the state vector and the original license plate corner coordinates obtained through the camera module (111) Vehicle pose estimation technology is used to compensate for license plate tilt caused by vehicle steering. Multispectral image data Through projection matrix Projecting the image onto the frontal plane yields a frontalized license plate image. .

6. The intelligent parking unattended vehicle access management system based on edge computing according to claim 5, characterized in that: The payment authentication unit (4) includes a blockchain smart contract module (41), a multi-factor authentication module (42), and a payment circuit breaker protection module (43). The blockchain smart contract module (41) is used to automatically execute parking fee settlement and transaction record uploading on the blockchain; The multi-factor authentication module (42) is used to combine license plate recognition, facial features and bioelectric signals to perform reliable authentication of user identity; The payment circuit breaker protection module (43) generates a comprehensive risk score by weighting multi-dimensional anomaly indicators. And automatically trigger the circuit breaker mechanism when abnormal payment behavior occurs.

7. A smart parking unattended vehicle access management method based on edge computing, based on the smart parking unattended vehicle access management system based on edge computing as described in any one of claims 1-6, characterized in that, Includes the following steps: S10.

1. Real-time environmental data is collected through a multimodal sensor array, processed by Kalman filtering and QBCN interference compensation algorithm, and the vehicle coordinates are output and the parking space occupancy status is dynamically updated. S10.

2. Based on multispectral imaging to capture multi-dimensional features of license plates, a lightweight CRNN model is used for cross-illumination scene recognition, and combined with vehicle pose estimation to achieve geometric correction and analysis of tilted and occluded license plates. S10.3 integrates a real-time rule engine and a spatiotemporal anomaly detection model, and allocates computing resources through a dynamic game optimization algorithm to achieve millisecond-level localized decision-making and system anomaly self-healing control. S10.

4. Relying on blockchain smart contracts to automatically execute dynamic fee calculation and multi-factor authentication, a tiered circuit breaker mechanism is triggered through transaction behavior analysis to ensure payment security and intercept abnormal transactions. S10.

5. Federated learning is used to aggregate the feature differences of multiple nodes, and spatiotemporal compression coding technology is combined to achieve efficient model updates and task scheduling.

Images