Traffic perception adjusting method and system for mobile vehicle-road cooperation roadside equipment

Abstract

The invention discloses a traffic perception adjusting method and system for mobile vehicle-road cooperation roadside equipment, and belongs to the field of intelligent traffic systems. The mobile vehicle-road cooperation roadside equipment traffic perception adjustment method comprises the following steps: collecting and preprocessing multi-source data of a traffic environment; performing feature extraction on the preprocessed multi-source data to obtain multi-source features; fusing the multi-source features by adopting an SAC model to generate a traffic perception result; performing online optimization and adjustment on parameters of the SAC model by using a quantum behavior particle swarm optimization algorithm; the optimized SAC model parameters are loaded, and rapid deployment and starting of the mobile roadside equipment are achieved. The system comprises mobile roadside equipment, a data preprocessing module, a feature extraction module, a data fusion adjustment module, a parameter optimization module, a rapid deployment module and the like. According to the method, the SAC and QPSO algorithms are combined, dynamic adjustment and optimization of the traffic perception model are realized, the perception accuracy and real-time performance are improved, and the method is suitable for various traffic scenes.

Description

Technical Field

[0001] The present invention relates to an intelligent transportation system, and in particular to a traffic perception adjustment method and system for mobile vehicle-road cooperative roadside equipment. Background Art

[0002] With the rapid development of intelligent transportation and vehicle-road cooperative technology, traffic perception systems play an important role in improving traffic safety, alleviating traffic congestion and realizing intelligent driving. However, the existing traffic perception systems have the following main problems: High complexity of data fusion: Multi-source heterogeneous data in traffic environments (such as cameras, lidar, millimeter-wave radar, etc.) require efficient fusion algorithms to obtain accurate perception results. Traditional methods have high computational complexity when processing large amounts of real-time data, making it difficult to provide accurate information in a timely manner.

[0003] Insufficient environmental adaptability: Traditional perception algorithms have unstable performance in different environments (such as weather changes, lighting conditions, and traffic flow changes), and are difficult to adapt to complex and changing traffic conditions, affecting the reliability and accuracy of the system.

[0004] Low deployment efficiency: The existing system takes a long time to deploy and debug in new traffic scenarios, requires a lot of parameter adjustments and model training, and cannot meet the needs of rapid response.

[0005] Lack of dynamic adjustment capabilities: Perception models are usually pre-trained and lack the ability to adjust and optimize based on real-time data during operation. When the environment changes, the model cannot adjust adaptively, resulting in a decrease in perception performance.

[0006] In order to solve the above problems, there is an urgent need for a traffic perception system that can be quickly deployed, has environmental adaptability and dynamic adjustment capabilities, can dynamically adapt to real-time changes in the traffic environment, optimize the perception model, and improve the system's perception accuracy and stability. Summary of the invention

[0007] Purpose of the invention: The purpose of the present invention is to provide a traffic perception adjustment method for mobile vehicle-road cooperative roadside equipment to solve the problems of high data fusion complexity, insufficient environmental adaptability, low deployment efficiency and lack of dynamic adjustment capability in existing traffic perception systems. Another purpose of the present invention is to propose a traffic perception adjustment system for mobile vehicle-road cooperative roadside equipment to solve the problem of how to execute the above adjustment method.

[0008] Technical solution: A traffic perception adjustment method for a mobile vehicle-road cooperative roadside device according to the present invention comprises the following steps: Collect multi-source data of traffic environment in real time through mobile roadside equipment and pre-process the multi-source data; Perform feature extraction on the preprocessed multi-source data to obtain multi-source features; The SAC model is used to fuse multi-source features to generate traffic perception results; The parameters of the SAC model are optimized and adjusted online using the quantum-behaved particle swarm optimization algorithm; Load the optimized SAC model parameters to achieve rapid deployment and startup of mobile roadside equipment.

[0009] Preferably, the mobile roadside equipment includes a camera, a lidar, a millimeter-wave radar, a V2X communication module and a high-performance edge computing unit.

[0010] Specifically, the mobile roadside equipment is used to be flexibly deployed in different traffic scenarios to collect multi-source heterogeneous data of the traffic environment. The equipment integrates multiple sensors such as cameras, lidars, and millimeter-wave radars to obtain real-time information about the road. The mobile roadside equipment is equipped with a high-performance edge computing unit with real-time data processing and algorithm operation capabilities. Through the V2X (Vehicle-to-Everything) communication module, the device realizes information interaction with vehicles and other roadside equipment, supports vehicle-road collaborative functions, and enhances the coverage and accuracy of traffic perception.

[0011] Preferably, the preprocessing of multi-source data includes cleaning, synchronizing and normalizing the multi-source data, wherein time synchronization uses a precision time protocol to achieve nanosecond-level synchronization.

[0012] Preferably, the feature extraction of the preprocessed multi-source data includes extracting high-level features from the preprocessed data using a deep learning model, wherein a convolutional neural network model is used to extract visual features from image data, and a PointNet++ model is used to extract spatial structure features from point cloud data; and motion features are extracted from radar data.

[0013] Preferably, the SAC model includes a strategy network and a value network, and the strategy network inputs multi-source features s t , output fusion strategy a t ; The value network evaluates the value function of the current strategy ; The use of the SAC model to fuse multi-source features includes: Multi-source features s t and fusion strategy a t Perform weighted fusion to obtain the fused features ; Using the fused features , generate traffic perception results.

[0014] Preferably, the policy network parameters of the SAC algorithm are The update goal is to maximize the following objective function: in, α is the temperature coefficient, D For the experience replay pool.

[0015] Preferably, the online optimization and adjustment of the parameters of the SAC model using the quantum behavior particle swarm optimization algorithm includes calculating the performance index of the SAC model using multi-source data collected in real time as a fitness function, updating the particle swarm, and optimizing the model parameters to achieve the best perception performance; The quantum behavior particle swarm optimization algorithm updates the position of the particle through the following position update formula: Furthermore, the above-mentioned mobile vehicle-road cooperative roadside equipment traffic perception adjustment method also includes the steps of: The traffic perception results are sent to nearby vehicles through the V2X communication interface to provide real-time traffic information and assist the vehicle's driving decisions. At the same time, the system receives the vehicle's status information to further enrich the perception data source and realize information interaction between mobile roadside equipment and vehicles and other roadside equipment.

[0016] The second aspect of the present invention discloses a mobile vehicle-road cooperative roadside equipment traffic perception adjustment system, comprising: Mobile roadside equipment, used to collect multi-source data of traffic environment in real time; A data preprocessing module, used for cleaning, synchronizing and normalizing the multi-source data collected by the mobile roadside equipment; The feature extraction module uses a deep learning model to extract features from the preprocessed data to obtain multi-source features; A data fusion adjustment module, based on the SAC algorithm, fuses the multi-source features to generate traffic perception results; Parameter optimization module, which uses quantum-behaved particle swarm optimization algorithm to optimize and online adjust the parameters of the SAC model to improve perception performance; Rapid deployment module, used to load optimized SAC model parameters before system deployment to achieve rapid deployment and startup of equipment; User interface module, used to provide operators with system status monitoring, traffic sensing result display, and parameter adjustment and control functions; The vehicle-road cooperative communication module is used to realize information interaction between mobile roadside equipment and vehicles and other roadside equipment.

[0017] Preferably, the vehicle-road cooperative communication module adopts the V2X communication protocol to achieve two-way transmission of traffic perception results and vehicle status information. The information is encoded in ASN.1 and complies with the V2X communication standard.

[0018] Based on the above technical solution, the data preprocessing module is used to clean, synchronize and normalize the multi-source data collected by the mobile roadside equipment. Since the data formats, sampling rates and timestamps of different sensors may differ, the data preprocessing module first performs format conversion and time synchronization on the data to ensure that the multi-source data is fused under the same time reference. Then, a filtering algorithm is used to remove noise and outliers to improve data quality. Normalization converts data of different scales and units to a unified range to provide consistent data input for subsequent feature extraction and data fusion.

[0019] Based on the above technical solution, the feature extraction module uses a deep learning model to extract high-level features from preprocessed data. For image data, a pre-trained convolutional neural network (CNN) is used to extract visual features, such as object detection and classification information; for point cloud data, a three-dimensional convolutional network or PointNet model is used to extract spatial structural features; for radar data, motion features such as the speed and distance of the target are extracted. The feature extraction module selects appropriate deep learning models and parameters for different types of data, extracts representative features, and provides a rich source of information for data fusion.

[0020] Based on the above technical solution, the data fusion adjustment module adopts a data fusion model based on the SAC (Soft Actor-Critic) algorithm to fuse multi-source features and generate unified traffic perception results. The SAC algorithm optimizes the data fusion strategy through the collaborative work of the policy network (Actor) and the value network (Critic). The policy network inputs multi-source features and outputs the optimal fusion strategy to generate traffic perception results such as target detection, tracking, and scene understanding. The value network evaluates the performance of the strategy and guides the update of the policy network. This module is capable of performing strategy optimization in high-dimensional continuous action space with high sample efficiency and stability.

[0021] Based on the above technical solution, the parameter optimization module uses the QPSO (quantum behavior particle swarm optimization) algorithm to optimize and online adjust the key parameters of the SAC model. The QPSO algorithm has global search capabilities and fast convergence characteristics by simulating quantum behavior. During the operation of the system, the parameter optimization module uses real-time collected data to calculate the performance indicators of the SAC model (such as perception accuracy, processing delay, etc.) as the fitness function. Then, the QPSO algorithm updates the particle swarm and optimizes the model parameters to achieve the optimal perception performance. This module realizes the dynamic adjustment of the perception model and enhances the adaptability of the system in different environments.

[0022] Based on the above technical solution, the quick deployment module is used to load the initial parameters of the SAC model that have been trained offline and optimized by QPSO before system deployment, so as to achieve rapid deployment and startup of the equipment. By pre-optimizing the model parameters and configuration, the on-site debugging and training time is reduced. The quick deployment module provides an automated deployment process, including network configuration of the equipment, sensor calibration and model loading, etc., supports one-click deployment function, and improves the deployment efficiency of the system.

[0023] Based on the above technical solution, the user interface module provides operators with an interface for system status monitoring, traffic perception result display and control operation. The user interface module displays the current traffic environment information, equipment operation status, working status of the perception model and traffic perception results through an intuitive graphical interface, ensuring that operators can grasp the operation of the system in real time. The module has parameter adjustment and control functions, and operators can adjust system parameters through the interface to meet specific application requirements. The user interface module also supports the query and analysis of historical data, helping operators optimize system configuration and improve the efficiency and accuracy of traffic perception.

[0024] Based on the above technical solution, the vehicle-road cooperative communication module realizes information interaction between mobile roadside equipment and vehicles and other roadside equipment. Through the V2X communication interface, the system sends traffic perception results to nearby vehicles, provides real-time traffic information, and assists vehicle driving decisions. At the same time, the system receives vehicle status information, such as location, speed, and driving direction, to further enrich the perception data source and improve the accuracy of perception.

[0025] The operation method of the above-mentioned mobile vehicle-road cooperative roadside equipment traffic perception and regulation system is as follows: Step 1: Data collection and preprocessing. The mobile roadside equipment collects multi-source data of the traffic environment in real time through the sensor module, including images, point clouds and radar data. The data preprocessing module cleans, synchronizes and normalizes the data to ensure the accuracy and real-time performance of the data. The processed data is transmitted to the feature extraction module.

[0026] Step 2: Feature extraction. The feature extraction module uses a deep learning model to extract high-level features from the preprocessed data, including visual features, spatial structure features, and motion features, providing rich information for data fusion.

[0027] Step 3: Data fusion and perception adjustment. The data fusion adjustment module uses a model based on the SAC algorithm to fuse multi-source features and generate traffic perception results. The parameter optimization module uses the QPSO algorithm to optimize and adjust the parameters of the SAC model online based on real-time data to improve the accuracy of data fusion and the environmental adaptability of the system.

[0028] Step 4: Quick deployment and startup. The quick deployment module loads the initial parameters of the SAC model that have been trained offline and optimized by QPSO before the system is deployed to achieve quick deployment and startup of the equipment. Through the automated deployment process, the on-site debugging and configuration time is reduced to ensure that the system can be put into operation quickly.

[0029] Step 5: Vehicle-road cooperative communication. The vehicle-road cooperative communication module sends the traffic perception results to nearby vehicles and other roadside equipment through the V2X communication interface to achieve information sharing. The system also receives vehicle status information, enriches the perception data source, and enhances the coverage of traffic perception.

[0030] Step 6: User interaction and system monitoring. The user interface module provides operators with an interface for system status monitoring, traffic sensing result display, and control operations. Operators can use the interface to understand the system operation status in real time, intervene and handle abnormal situations, and adjust system parameters to improve the operational flexibility of the system.

[0031] Step 7: Online parameter optimization and exception handling. During operation, the system continuously optimizes model parameters online to adapt to environmental changes. When an abnormal situation is detected (such as sensor failure, decreased perception accuracy, etc.), the system automatically diagnoses and handles the fault, including restarting the device, switching to a backup module, or sending an alarm message to ensure stable operation of the system.

[0032] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: (1) High-precision traffic perception: Through the fusion of multi-source data and the application of deep learning models, high-precision perception of the traffic environment is achieved, including the detection and tracking of vehicles, pedestrians, and non-motor vehicles.

[0033] (2) Strong real-time performance: The use of high-performance edge computing units and optimized algorithm structures ensures the real-time performance of the system and meets the timeliness requirements of traffic perception.

[0034] (3) Strong environmental adaptability: Using the SAC and QPSO algorithms, the system can dynamically adjust model parameters according to different traffic environments and weather conditions to maintain optimal performance.

[0035] (4) Rapid deployment and easy maintenance: The rapid deployment of modules enables rapid installation and startup of equipment, reducing on-site debugging time; the modular design of the system facilitates maintenance and upgrades.

[0036] (5) Vehicle-road collaboration: Through the vehicle-road collaboration communication module, information interaction with vehicles is realized, the overall perception capability of the system is enhanced, and traffic safety and efficiency are improved.

[0037] (6) User-friendliness: The user interface module provides an intuitive interface, which makes it easy for operators to monitor and control the system, thus improving the system’s usability. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a schematic diagram of the overall structure of the traffic sensing and regulation system; Figure 2 This is a schematic diagram of the hardware structure; Figure 3 This is the flow chart of data acquisition and preprocessing module; Figure 4 It is the workflow diagram of the combination of SAC and QPSO algorithm; Figure 5 A system deployment flow chart for the rapid deployment module; Figure 6 Schematic diagram of the user interaction and system monitoring interface. DETAILED DESCRIPTION

[0039] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0040] Example 1: Figure 1 and Figure 2 As shown, this embodiment provides a mobile vehicle-road cooperative roadside equipment traffic perception adjustment system. The system mainly includes mobile roadside equipment, data preprocessing module, feature extraction module, data fusion adjustment module, parameter optimization module, rapid deployment module, user interface module and vehicle-road cooperative communication module.

[0041] Mobile roadside equipment: used for flexible deployment in different traffic scenarios to collect multi-source heterogeneous data of the traffic environment. The equipment integrates cameras, lidar, millimeter-wave radar, V2X communication modules and high-performance edge computing units.

[0042] Camera: It uses an industrial-grade high-resolution camera with high sensitivity and wide dynamic range, which can collect clear image data under complex lighting conditions such as strong light and weak light. The camera's resolution is not less than 1920×1080, and the frame rate can reach 60 frames per second, meeting the needs of real-time video acquisition. Laser radar: It uses multi-beam laser radar, such as 32-line or 64-line laser radar, which has wide-angle and high-precision spatial scanning capabilities. The scanning frequency of laser radar can reach more than 10Hz, and the ranging accuracy is better than ±2cm, which can obtain high-precision three-dimensional point cloud data of the surrounding environment.

[0043] Millimeter-wave radar: The operating frequency band is 76GHz~77GHz, with strong anti-interference ability and little influence from weather. The detection distance of millimeter-wave radar can reach more than 200 meters, and the speed resolution is better than 0.1m / s, which can accurately measure the speed, distance and angle of the target object.

[0044] V2X communication module: supports DSRC (Dedicated Short Range Communications) or LTE-V2X (Cellular Vehicle-to-Everything) communication protocols to achieve information exchange between devices and vehicles. The communication distance can reach more than 1,000 meters, with low latency and high reliability.

[0045] High-performance edge computing unit: Adopts hardware acceleration platforms such as embedded GPU or FPGA, such as NVIDIA JetsonAGX Xavier, which has powerful computing power and AI reasoning performance. It supports real-time reasoning of deep learning models and the operation of complex algorithms, meeting the high real-time and high-precision requirements of traffic perception.

[0046] Data preprocessing module: cleans, synchronizes and normalizes multi-source heterogeneous data from mobile roadside equipment to ensure the accuracy and consistency of subsequent processing.

[0047] Data cleaning: For data from different sensors, corresponding filtering and denoising algorithms are used. For example, for image data, median filtering and Gaussian filtering are used to remove noise; for lidar point cloud data, statistical filtering and conditional filtering are used to remove outliers; for millimeter-wave radar data, Kalman filters are used to eliminate measurement noise.

[0048] Data synchronization: Since the data sampling frequencies and timestamps of different sensors may differ, a time synchronization algorithm is used to align the data of different sensors to a unified time base. Specifically, the IEEE 1588 Precision Time Protocol (PTP) is used to achieve nanosecond-level time synchronization, or a hardware trigger mechanism is used to synchronize the sampling moments of each sensor.

[0049] Data normalization: Normalize data of different scales and units to ensure that the data is integrated at the same scale. For numerical features, use the Z-score normalization method to convert the data to [0,1] or the standard normal distribution range; for categorical features, use one-hot encoding or embedding representation.

[0050] Feature extraction module: Use deep learning models to extract features from preprocessed data to obtain high-level semantic information.

[0051] Image feature extraction: Use a pre-trained convolutional neural network (CNN) model to detect and classify objects in image data. Through multi-layer convolution and pooling operations, the spatial features and semantic information of the image are extracted. The specific process is as follows: Input image: preprocessed image data, size .

[0052] Convolutional layer extracts features: Through a series of convolutional layers and activation functions, feature maps are extracted , where C is the number of channels.

[0053] Pooling layer dimensionality reduction: Use maximum pooling or average pooling layer to reduce the size of the feature map and enhance the translation invariance of the model.

[0054] Fully connected layer: Flatten the feature map, input it into the fully connected layer, and directly output the category and position of the target.

[0055] Point cloud feature extraction: Use the PointNet++ model to extract features from the three-dimensional point cloud data of the LiDAR. The specific process is as follows: Input point cloud data: preprocessed point cloud data, containing N points, the coordinates of each point are .

[0056] Point-by-point feature learning: Through a multi-layer perceptron (MLP), the coordinates of each point are nonlinearly mapped to extract point-by-point features .

[0057] Global feature aggregation: Use the maximum pooling layer to aggregate the features of all points into global features .

[0058] Classification and regression: Use global features to classify and estimate the location of targets.

[0059] Radar feature extraction: Extract the speed of the target object from the millimeter wave radar data ,distance and azimuth , forming the target's motion feature vector The radar echo signal is processed using signal processing algorithms and target tracking algorithms, such as constant false alarm rate (CFAR) detection and Kalman filtering.

[0060] Data fusion adjustment module: It adopts a model based on the SAC (Soft Actor-Critic) algorithm to fuse multi-source features and generate unified traffic perception results.

[0061] Principle of SAC algorithm: SAC algorithm is a deep reinforcement learning algorithm based on maximum entropy, which aims to maximize the weighted sum of the expected return of the strategy and the strategy entropy, and improve the exploration and stability of the strategy.

[0062] Policy Network (Actor): Input Multi-Source Features , output fusion strategy , which is the weight coefficient of each feature. The policy network parameters are , implemented using a deep neural network. The network structure is as follows: Input layer: receiving multi-source features .

[0063] Hidden layer: multi-layer fully connected layer, the activation function is ReLU or Tanh.

[0064] Output layer: output fusion strategy , using the Softmax function to normalize the weights to the range of [0,1].

[0065] Value Network (Critic): The value function that evaluates the current strategy , the parameters are The structure of the value network is similar to that of the policy network, but the output is a scalar value.

[0066] Loss function: The goal of the policy network is to maximize the following objective function: Among them, α is the temperature coefficient, which controls the randomness of the strategy; D is the experience replay pool.

[0067] The loss function of the value network is: Where γ is the discount factor, is the target value function.

[0068] Data fusion process: Feature fusion: multi-source features and fusion strategy Perform weighted fusion to obtain the fused features : in, Represents bit-wise multiplication of elements.

[0069] Perception result generation: using fused features , through fully connected layers or other neural network structures, traffic perception results such as target detection, classification and positioning are generated.

[0070] Parameter optimization module: Use the QPSO (quantum-behaved particle swarm optimization) algorithm to optimize and online adjust the key parameters of the SAC model.

[0071] Principle of QPSO algorithm: QPSO algorithm updates the position of particles by simulating the quantum behavior of particles, and has global search capabilities and fast convergence characteristics.

[0072] Particle position update formula: Fitness function: The performance indicators of the SAC model are used as the fitness function, such as perception accuracy Acc and processing delay The weighted sum of: in, is the position of particle i, i.e., the parameter of the SAC model; W1 and W2 are weight coefficients, reflecting the importance of perception accuracy and processing delay.

[0073] Rapid deployment module: used to load the initial parameters of the SAC model that have been trained offline and optimized by QPSO before system deployment, so as to achieve rapid deployment and startup of the equipment.

[0074] Deployment process: Offline model training: Based on historical traffic data, train the SAC model to obtain initial parameters and .

[0075] QPSO parameter optimization: Use the QPSO algorithm to optimize the SAC model parameters and obtain the optimized parameters and .

[0076] Model parameter loading: Load the optimized model parameters to the edge computing unit.

[0077] Network configuration: Set the network connection parameters of the device, including IP address, communication protocol, etc.

[0078] Sensor calibration: Calibrate cameras, lidar, and millimeter-wave radar to ensure the accuracy of data collection.

[0079] System startup: Start the system and enter real-time operation state.

[0080] User interface module: provides operators with system status monitoring, traffic perception result display, and parameter adjustment and control functions.

[0081] System status monitoring: Real-time display of sensor status, network connection status, edge computing unit operation status, etc. Including CPU, GPU utilization, memory usage, sensor connection status, etc.

[0082] Traffic perception display: Graphically display vehicle detection, trajectory tracking, traffic flow statistics and other information. It can display real-time video images and overlay detected target boxes and trajectory lines.

[0083] Parameter adjustment and control: Provides an adjustment interface for model parameters, allowing operators to adjust system parameters as needed, such as model weights, learning rates, thresholds, etc.

[0084] Abnormal alarm: When the system detects an abnormal situation, such as sensor failure, network interruption, model abnormality, etc., it issues an alarm message to prompt the operator to handle it.

[0085] Vehicle-road cooperative communication module: realizes information interaction between mobile roadside equipment and vehicles and other roadside equipment.

[0086] Communication protocol: Adopt standard V2X communication protocols, such as IEEE 802.11p or LTE-V2X, to ensure compatibility with vehicles and communication quality.

[0087] Information interaction content: Send information: Send traffic perception results, such as road conditions ahead, congestion information, accident warnings, etc., to nearby vehicles to assist driving decisions.

[0088] Receive information: Receive vehicle status information, such as location ,speed , Driving direction Etc., to further enrich the perception data source.

[0089] Data format: Information is encoded using ASN.1, which complies with the V2X communication standard and ensures the standardization and reliability of data transmission.

[0090] Based on the above hardware and software modules, dynamic optimization of traffic perception adjustment is achieved. The system operation method includes the following steps: Step 1: Data collection and preprocessing Mobile roadside equipment collects multi-source data of the traffic environment in real time through cameras, lidar, millimeter-wave radar and V2X communication modules. The data preprocessing module cleans, synchronizes and normalizes the collected data to ensure the accuracy and consistency of the data.

[0091] Step 2: Feature Extraction The feature extraction module uses deep learning models to extract high-level features from preprocessed data. For image data, it extracts visual features; for point cloud data, it extracts spatial structure features; for radar data, it extracts motion features.

[0092] Step 3: Data Fusion and Perception Adjustment The data fusion and adjustment module uses the SAC algorithm to fuse multi-source features and generate traffic perception results. The parameter optimization module uses the QPSO algorithm to optimize and adjust the parameters of the SAC model online based on real-time data to improve the accuracy of data fusion and the environmental adaptability of the system.

[0093] Step 4: Quick deployment and startup Before system deployment, the rapid deployment module loads the initial parameters of the SAC model that have been trained offline and optimized by QPSO to achieve rapid deployment and startup of the equipment. Through the automated deployment process, the on-site debugging and configuration time is reduced to ensure that the system can be put into operation quickly.

[0094] Step 5: Vehicle-Road Collaborative Communication The vehicle-road cooperative communication module sends traffic perception results to nearby vehicles and other roadside equipment through the V2X communication interface to achieve information sharing. The system also receives vehicle status information, enriches the perception data source, and enhances the coverage of traffic perception.

[0095] Step 6: User Interaction and System Monitoring The user interface module provides operators with system status monitoring, traffic perception result display, and parameter adjustment and control functions. Operators can understand the system operation status in real time through the interface, intervene and handle abnormal situations, and adjust system parameters to improve the operational flexibility of the system.

[0096] Step 7: Online parameter optimization and exception handling During the operation of the system, the parameter optimization module continuously optimizes the model parameters online to adapt to environmental changes. When an abnormal situation is detected (such as sensor failure, reduced perception accuracy, etc.), the system automatically diagnoses and handles the fault, including restarting the device, switching to a backup module, or sending an alarm message to ensure the stable operation of the system.

[0097] Example 2: This example introduces the application of the system in a highway scenario, such as Figure 3In the highway scenario, vehicles travel at high speeds and the traffic environment is relatively simple, but the system is required to have high real-time and high-precision perception capabilities.

[0098] Step 1: Data collection and preprocessing Mobile roadside equipment is deployed on the side of the highway. The camera frame rate is set to 60 frames per second, the laser radar scanning frequency is 20Hz, and the millimeter wave radar refresh rate is 50Hz. The data preprocessing module uses parallel processing and multi-threading technology for high-speed data streams to improve data processing efficiency.

[0099] Step 2: Feature Extraction The feature extraction module adopts a deep learning model to improve processing speed and meet real-time requirements.

[0100] Step 3: Data Fusion and Perception Adjustment The data fusion adjustment module adjusts the parameters of the SAC model, simplifies the network structure, and reduces the amount of calculation. The speed error term is added to the loss function. , improve the accuracy of vehicle speed estimation: Step 4: Parameter Optimization The parameter optimization module monitors perception performance indicators in real time, such as detection accuracy and processing delay. Using the QPSO algorithm, the parameters of the SAC model, such as learning rate, number of network layers, number of neurons, etc., are dynamically adjusted.

[0101] Step 5: Vehicle-Road Collaborative Communication The vehicle-road cooperative communication module sends the vehicle information about the road ahead, accident warnings, etc., to help the vehicle make decisions in advance. The communication frequency is appropriately reduced to save communication resources.

[0102] Step 6: User Interaction The user interface module provides traffic management personnel with real-time traffic conditions on highway sections, including traffic flow statistics, average vehicle speed and other information.

[0103] Example 3: This example introduces the application of the system in urban road scenarios, such as Figure 4 , Figure 5 and Figure 6 The urban road traffic environment is complex, with multiple traffic participants such as vehicles, pedestrians, and bicycles mixed together, which places higher requirements on the system's perception ability.

[0104] Step 1: Data collection and preprocessing Mobile roadside equipment is deployed at the intersection of urban roads. Cameras, lidar and millimeter-wave radar work together to collect dense traffic data. The data preprocessing module uses multi-threading and caching mechanisms to prevent data loss.

[0105] Step 2: Feature Extraction The feature extraction module uses a deep neural network to extract high-precision features. For multi-category targets, a multi-task learning model is used to simultaneously detect vehicles, pedestrians, and non-motor vehicles.

[0106] Step 3: Data Fusion and Perception Adjustment The data fusion adjustment module uses a multi-task learning method to handle the detection tasks of different types of targets. The loss function is: in, is the loss of the ith task, is the corresponding weight coefficient.

[0107] Step 4: Parameter Optimization The parameter optimization module uses the QPSO algorithm to optimize a large number of model parameters. Introducing adaptive inertia weights : Step 5: Vehicle-Road Collaborative Communication The vehicle-road cooperative communication module adopts multi-access edge computing (MEC) technology to disperse the communication load and improve communication efficiency.

[0108] Step 6: User Interaction The user interface module provides traffic management departments with rich traffic data analysis functions, including congestion prediction, traffic signal optimization suggestions, etc.

[0109] Embodiment 4: This embodiment introduces the application of the system under adverse weather conditions, such as rainy days, snowy days, foggy days, etc. Adverse weather poses challenges to sensor performance and perception algorithms.

[0110] Step 1: Data collection and preprocessing In bad weather conditions, the camera's image quality deteriorates. The system relies more on the data from LiDAR and millimeter-wave radar. The data preprocessing module strengthens the denoising of sensor data and filters out noise and interference caused by weather.

[0111] Step 2: Feature Extraction For low-quality image data, the parameters of the deep learning model are adjusted to enhance the processing capabilities of blurred and low-contrast images. For point cloud data, a robust feature extraction algorithm is used to reduce the impact of rain and snow on LiDAR.

[0112] Step 3: Data Fusion and Perception Adjustment The data fusion adjustment module automatically adjusts the weight of each sensor data and increases the reliance on lidar and millimeter-wave radar data. The SAC algorithm dynamically adjusts the fusion strategy based on the feedback of the perception results.

[0113] Step 4: Parameter Optimization The parameter optimization module uses the QPSO algorithm to optimize model parameters in real time and improve perception accuracy. A robustness index is added to the fitness function to evaluate the performance of the model under harsh conditions.

[0114] In summary, the present invention provides a traffic perception and regulation system based on mobile vehicle-road cooperative roadside equipment and its operation method, which makes full use of multi-source sensor data and advanced deep learning and optimization algorithms to achieve high accuracy, real-time and environmental adaptability of traffic perception. The system has the advantages of rapid deployment, intelligent regulation and user-friendliness, and has broad application prospects.

Claims

1. A traffic perception adjustment method for mobile vehicle-road cooperative roadside equipment, characterized in that: The steps include: Collect multi-source data of traffic environment in real time through mobile roadside equipment and pre-process the multi-source data; Perform feature extraction on the preprocessed multi-source data to obtain multi-source features; The SAC model is used to fuse multi-source features to generate traffic perception results; The parameters of the SAC model are optimized and adjusted online using the quantum-behaved particle swarm optimization algorithm; Load the optimized SAC model parameters to achieve rapid deployment and startup of mobile roadside equipment.

2. The traffic perception adjustment method for mobile vehicle-road cooperative roadside equipment according to claim 1 is characterized in that: The mobile roadside equipment includes a camera, a laser radar, a millimeter wave radar, a V2X communication module and a high-performance edge computing unit.

3. The traffic perception adjustment method for mobile vehicle-road cooperative roadside equipment according to claim 1 is characterized in that: The preprocessing of multi-source data includes cleaning, synchronizing and normalizing the multi-source data, wherein time synchronization uses a precision time protocol to achieve nanosecond-level synchronization.

4. The traffic perception adjustment method for mobile vehicle-road cooperative roadside equipment according to claim 1 is characterized in that: The feature extraction of the preprocessed multi-source data includes extracting high-level features from the preprocessed data using a deep learning model, wherein a convolutional neural network model is used to extract visual features from image data, and a PointNet++ model is used to extract spatial structure features from point cloud data; and motion features are extracted from radar data.

5. The traffic perception adjustment method for mobile vehicle-road cooperative roadside equipment according to claim 1 is characterized in that: The SAC model includes a policy network and a value network. The policy network inputs multi-source features. s t , output fusion strategy a t ; The value network evaluates the value function of the current strategy ; The use of the SAC model to fuse multi-source features includes: Multi-source features s t and fusion strategy a t Perform weighted fusion to obtain the fused features ; Using the fused features , generate traffic perception results.

6. The traffic perception adjustment method for mobile vehicle-road cooperative roadside equipment according to claim 5 is characterized in that: Policy network parameters of the SAC algorithm The update goal is to maximize the following objective function: in, α is the temperature coefficient, D For the experience replay pool.

7. The traffic perception adjustment method for mobile vehicle-road cooperative roadside equipment according to claim 1 is characterized in that: The online optimization and adjustment of the parameters of the SAC model using the quantum behavior particle swarm optimization algorithm includes calculating the performance index of the SAC model as a fitness function using multi-source data collected in real time, updating the particle swarm, and optimizing the model parameters to achieve the best perception performance; The quantum behavior particle swarm optimization algorithm updates the position of the particle through the following position update formula: in, For the Substitute the position of the ith particle; For the The historical optimal position of the i-th particle; For the The global optimal position; is the contraction-expansion factor; is a random number between [0,1].

8. The traffic perception adjustment method for mobile vehicle-road cooperative roadside equipment according to claim 1 is characterized in that: Also includes the steps: The traffic perception results are sent to nearby vehicles through the V2X communication interface to provide real-time traffic information and assist the vehicle's driving decisions. At the same time, the system receives the vehicle's status information to further enrich the perception data source and realize information interaction between mobile roadside equipment and vehicles and other roadside equipment.

9. A mobile vehicle-road cooperative roadside equipment traffic perception and adjustment system, characterized in that: include: Mobile roadside equipment, used to collect multi-source data of traffic environment in real time; A data preprocessing module, used for cleaning, synchronizing and normalizing the multi-source data collected by the mobile roadside equipment; The feature extraction module uses a deep learning model to extract features from the preprocessed data to obtain multi-source features; A data fusion adjustment module, based on the SAC algorithm, fuses the multi-source features to generate traffic perception results; Parameter optimization module, which uses quantum-behaved particle swarm optimization algorithm to optimize and online adjust the parameters of the SAC model to improve perception performance; Rapid deployment module, used to load optimized SAC model parameters before system deployment to achieve rapid deployment and startup of equipment; User interface module, used to provide operators with system status monitoring, traffic sensing result display, and parameter adjustment and control functions; The vehicle-road cooperative communication module is used to realize information interaction between mobile roadside equipment and vehicles and other roadside equipment.

10. The mobile vehicle-road cooperative roadside equipment traffic perception and adjustment system according to claim 9 is characterized in that: The vehicle-road cooperative communication module adopts the V2X communication protocol to realize two-way transmission of traffic perception results and vehicle status information. The information is encoded in ASN.1 and complies with the V2X communication standard.

Images