Vehicle-road cooperative system based on 4D millimeter-wave radar

Abstract

A vehicle-road cooperative system based on 4D millimeter-wave radar includes: a 4D millimeter-wave radar system unit, a cloud server unit, a road traffic information service unit and a user terminal unit; the 4D millimeter-wave radar system unit transmits electromagnetic waves and receives target echoes, obtains vehicle information and road environment information, and generates road data; the cloud server unit receives and stores road data, and serves as a platform for processing road data; the road traffic information service unit identifies traffic accidents, extracts road features, and identifies abnormal conditions and dangerous driving vehicles on the road to obtain traffic road information data; the user terminal unit receives traffic road information data, converts it into a 3D model to display to the driver, and provides auxiliary information for the automatic driving system. The present invention can provide comprehensive road information for relevant departments for road cooperative management, and can provide auxiliary driving data for drivers and automatic driving systems in real time to improve driving safety.

Description

Technical Field

[0001] The present invention belongs to the fields of millimeter-wave radar detection, radar signal processing, and radar application technology, as well as intelligent transportation technology. In particular, it relates to a vehicle-road cooperative system based on 4D millimeter-wave radar, which realizes road information acquisition and monitoring based on millimeter waves, and realizes vehicle-road cooperation. Background Art

[0002] As the number of vehicles increases, traffic congestion and accidents become more frequent. In the future, autonomous vehicles will become the mainstream mode of transportation. New road service products are needed to provide comprehensive and integrated traffic and road information, ensuring effective control of the driver and the autonomous driving system, and ensuring real-time road management by relevant departments. Vehicle-road cooperative systems utilize roadside sensors to capture road environment data and provide auxiliary information to the autonomous driving system to ensure safety. They also provide detailed traffic information to road management departments for real-time traffic control and to alleviate congestion.

[0003] Existing road test sensors used in vehicle-road cooperative technology include lidar, millimeter-wave radar, and camera sensors. Lidar is a new type of radar sensor whose transmitted and received signals are short-wavelength lasers. This radar has a long detection range and high accuracy, and can image the specific outlines of detected objects. However, it is relatively expensive, and due to the weak diffraction ability of laser signals, lidar's detection accuracy is significantly affected by inclement weather, strong light interference, or obstructions. Camera sensors are relatively low-cost, and the information processing technology used to generate images based on cameras is relatively mature. However, these sensors have the disadvantages of large data volumes and severe environmental interference. Millimeter-wave radar is a radar sensor operating in the millimeter-wave frequency band. Traditional millimeter-wave radar sensors are relatively low-cost and can detect the speed, distance, and azimuth of a target. Because millimeter-wave electromagnetic waves have strong penetration capabilities in rain, fog, and dust, this radar has the ability to operate in all weather conditions. However, its detection accuracy is lower than that of lidar, and it cannot image the outline of a target or static environments.

[0004] The limitations of single sensor detection mentioned above can be overcome through multi-sensor fusion detection. However, due to differences in the working principles of sensors, it is difficult to ensure the accuracy of the information obtained in the time and space domains by fusing different types of sensor data. Therefore, it is necessary to adopt a new roadside sensor solution and customize its data to obtain comprehensive, high-precision, and all-weather road traffic service information. Summary of the Invention

[0005] In order to overcome the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide a vehicle-road cooperative system based on 4D millimeter-wave radar to ensure that relevant departments can manage roads more effectively in real time and provide more comprehensive road information for drivers and self-driving cars.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is:

[0007] A vehicle-road cooperative system based on 4D millimeter-wave radar, including:

[0008] A 4D millimeter-wave radar system unit is used to acquire vehicle information and road environment information to generate road data; the vehicle information includes the speed, distance, angle, and altitude of vehicles on the road; the road environment information includes the distance, angle, and altitude information acquired by the radar of various static targets on the road; the road data includes vehicle information, vehicle trajectory, and a point cloud image of the road environment;

[0009] The cloud server unit receives and stores all road data sent by the 4D millimeter-wave radar system unit and serves as a platform for processing road data;

[0010] The road traffic information service unit identifies traffic accidents, extracts road features, and identifies abnormal conditions and dangerously driven vehicles on the road based on road data stored in the cloud server to obtain traffic road information data. The road features include traffic volume, queue length, and average speed. The traffic road information data includes dangerous driving data on any road section, road anomaly data, vehicle trajectory information, road environment point cloud images, and traffic accident data.

[0011] The user terminal unit receives traffic road information data transmitted by the cloud server, converts it into a 3D model and displays it to the driver, and provides auxiliary information for the autonomous driving system.

[0012] In one embodiment, the 4D millimeter wave radar system unit includes:

[0013] Antenna unit, which transmits electromagnetic waves and receives target echoes to obtain raw data of the vehicle and road environment;

[0014] a signal processing unit, processing the raw data to obtain a data set containing vehicle information and road environment information;

[0015] a speed and distance extraction unit, for obtaining the speed and distance of the vehicle from the data set based on a CFAR detection algorithm;

[0016] An angle calculation unit, based on a DOA detection algorithm, obtains the angle of the vehicle from the data set; the angle of the vehicle includes its pitch angle and azimuth angle;

[0017] The target tracking unit uses the Kalman filter algorithm and track initiation algorithm to track multiple targets and generate vehicle trajectories based on the vehicle's speed, distance and angle;

[0018] The environmental perception unit processes the road environment information in the data set based on the point cloud imaging algorithm and constructs a three-dimensional image of the road environment, namely the road environment point cloud image;

[0019] An information output unit outputs the vehicle's speed, distance, angle, vehicle trajectory, and road environment point cloud image to a cloud server unit;

[0020] In one embodiment, the road traffic information service unit includes:

[0021] A traffic information acquisition unit, which acquires the road data;

[0022] The traffic accident identification unit processes the road environment point cloud image using a traffic accident identification and classification algorithm based on a generalized regression neural network to identify the traffic accident, labels the area where the traffic accident occurred, and transmits the traffic accident information to the data integration and transmission unit; the traffic accident information includes: the road environment point cloud image with the traffic accident area labeled;

[0023] The dangerous driving detection unit matches vehicle trajectories with road environment point cloud images in time and space to generate a complete traffic image. It then processes the matched traffic image using a deep learning algorithm based on a generalized regression neural network. By establishing a dangerous driving recognition model, it labels target vehicles involved in dangerous driving and their specific violations, and transmits the dangerous driving data to the data integration and transmission unit.

[0024] The road anomaly detection unit uses the YOLOv3-based image detection algorithm to process road environment point cloud images, trains a program to identify anomalies in the road environment point cloud images, labels the road anomaly areas and their possible anomaly states, and transmits the road anomaly data to the data integration and transmission unit;

[0025] The road feature extraction unit uses the vehicle speed, distance and vehicle trajectory to obtain the traffic volume, queue length and average road speed on the road as road features through data analysis, and transmits the road feature data to the data integration and transmission unit;

[0026] Data integration and sending unit: summarizes traffic accident data, dangerous driving data, road anomaly data, road feature data, as well as road vehicle information, vehicle trajectory information and road environment data by time, stores and uploads them to the traffic management department for query at any time, and transmits dangerous driving data, road anomaly data, vehicle trajectory information, road environment data to the user terminal unit to provide useful road information for drivers and autonomous driving systems.

[0027] In one embodiment, the traffic accident classification algorithm includes:

[0028] Firstly, the spatial first-order and second-order partial derivatives of the acquired road environment point cloud image are extracted as features for traffic accident discrimination and classification. The features are used to determine the input nodes of the generalized regression neural network and initialize the neural network.

[0029] Secondly, images containing various types of traffic accidents are selected from the stored road environment point cloud image library to form training samples for road anomaly detection. The generalized regression neural network is trained and the actual output of the network and the state of neurons in each layer are determined.

[0030] Then, the gradient descent algorithm is used to determine the weights and thresholds between the hidden layer and the output layer of the generalized regression neural network. The weights and thresholds are updated according to the output error to correct the nonlinear transfer function inside the hidden layer nodes of the generalized neural network.

[0031] Finally, the error of traffic accident discrimination and classification is calculated. If the traffic accident discrimination and classification error meets the actual application requirements, the generalized regression neural network terminates learning. Otherwise, the image is reselected for learning. At this point, the construction of the traffic accident discrimination classifier is completed, and traffic accidents are discriminated based on the constructed traffic accident discrimination classifier.

[0032] In one embodiment, the dangerous driving recognition model can identify dangerous driving behaviors including speeding, forced lane change, forced overtaking, and driving on occupied lanes; the road anomalies include obstacles, falling rocks, collapses, and damage.

[0033] In one embodiment, the road anomaly data is identified by a point cloud image of the road environment that indicates an abnormal condition, and the method is as follows:

[0034] The stored road environment point cloud image is used as a data set, and features are extracted from the data set through a basic feature extraction network to obtain a feature map;

[0035] A grid of the same size as the feature map is set on the input image. If the center coordinates of the target to be detected, i.e., the road anomaly, fall within this grid, it will be predicted by this grid.

[0036] Three associated boxes are predicted for each grid, and the associated box most relevant to the target to be detected is finally selected to mark the target.

[0037] In one embodiment, the user terminal unit comprises:

[0038] Information receiving unit: receives processed information from the road traffic information service unit, including: dangerous driving data of any road section, road anomaly data, vehicle trajectory information, road environment point cloud images, and traffic accident data;

[0039] 3D imaging unit: matches, calculates, and fits the received traffic accident data, dangerous driving data, road anomaly data, vehicle trajectory information, and road environment data to establish a 3D image model. Ultimately, the model of the vehicle on the road, the road environment model, and the dangerous driving warning model are transmitted to the vehicle terminal for display and then to the autonomous driving assistance unit.

[0040] Autonomous driving assistance unit: processes the received model data into a data format that can be processed by the autonomous driving system, transmits it to the autonomous driving system through the terminal, and combines it with the information independently obtained by the autonomous driving system to assist the autonomous driving system.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] (1) Compared with conventional road traffic radars, the 4D millimeter wave radar system involved in the present invention can detect vehicle target height information and perform point cloud imaging of the road environment at a lower cost and with higher accuracy in obtaining road information.

[0043] (2) The road traffic information service unit involved in the present invention can process radar data to extract comprehensive road information and upload all road data to relevant departments to ensure their coordinated management of roads.

[0044] (3) The vehicle-road cooperative system involved in the present invention can perform 3D modeling on the road status data obtained by the road traffic information service unit, display it to the driver in real time to help him make timely judgments while driving, and transmit it to the automatic driving system as auxiliary information to improve the safety of automatic driving. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is an implementation flow chart of the present invention.

[0046] Figure 2 It is a system structure diagram of the present invention.

[0047] Figure 3 Schematic diagram of the system application of the present invention.

[0048] Figure 4 This is a block diagram of the millimeter wave radar system unit in the system of the present invention.

[0049] Figure 5 It is a block diagram of the cloud server unit in the system of the present invention.

[0050] Figure 6 It is a block diagram of the road traffic information service unit in the system of the present invention.

[0051] Figure 7 It is a block diagram of a user terminal unit in the system of the present invention. DETAILED DESCRIPTION

[0052] The embodiments of the present invention are described in detail below with reference to the accompanying drawings and examples.

[0053] The present invention is a vehicle-road cooperative system based on 4D millimeter-wave radar. It introduces vehicle height information based on the 4D millimeter-wave radar and performs point cloud imaging of the road environment to realize vehicle-road cooperative management.

[0054] 4D millimeter-wave radar is a new type of sensor developed based on the millimeter-wave radar sensor. Compared with traditional millimeter-wave radar, it can perform high-precision detection of the target's speed, direction, height, and distance in four dimensions. The radar detection data can be used to image the contours, categories, behaviors, and static environments of moving targets. This means that the radar can meet the requirements of the vehicle-road cooperative system for roadside sensor performance, including the recognition of smaller objects, partially obscured objects, and the detection of stationary objects and laterally moving obstacles. It can detect various information on the road with higher accuracy and resolution, all day and night.

[0055] The present invention uses 4D millimeter-wave radar to acquire data and process it to obtain detailed traffic information (such as road flow ahead, queue length, dangerous driving information, road anomaly information, accident scene information, etc.) for relevant departments to conduct traffic regulation and alleviate road congestion; at the same time, the 3D modeled road scene information is processed and displayed to the driver in real time to help him make timely judgments while driving, and can be transmitted to the automatic driving system as auxiliary information to improve the safety of automatic driving.

[0056] Specifically, if Figure 1 and Figure 2 As shown, the vehicle-road cooperative system of the present invention mainly includes a 4D millimeter-wave radar system unit, a cloud server unit, a road traffic information service unit and a user terminal unit.

[0057] Reference Figure 3 In the application scenario of the system of the present invention, 4D millimeter-wave radars installed on both sides of the road acquire information about vehicles and the road environment, obtaining target information and road environment information, and transmit this information to a cloud server unit. A road traffic information service unit running on the cloud server processes the received information to obtain traffic accident data, road characteristic data, road anomaly data, and dangerous driving data. All traffic and road information is uploaded to relevant departments for traffic management, and traffic and road information that can assist driving is transmitted to user terminals installed in vehicles.

[0058] The 4D millimeter-wave radar system unit of the present invention is primarily used to acquire vehicle information and road environment information, generating road data. In this invention, vehicle information includes the speed, distance, angle, and altitude of vehicles on the road. Road environment information includes the distance, angle, and altitude information acquired by the radar for various static targets on the road. Road data includes vehicle information, vehicle trajectories, and point cloud images of the road environment.

[0059] Reference Figure 4 The 4D millimeter-wave radar system unit of the present invention mainly includes: an antenna unit, a signal processing unit, a speed and distance extraction unit, an angle calculation unit, a target tracking unit, an environment perception unit and an information output unit. The antenna unit transmits electromagnetic waves and receives target echoes to obtain the original data of road vehicles and road environment; the original data is processed by the signal processing unit to generate a data set containing vehicle information and road environment information; the speed and distance extraction unit uses the CFAR detection algorithm to obtain the speed and distance of the vehicle target from the data set, and transmits the obtained target speed and distance information to the target tracking unit and the information output unit; the angle calculation unit obtains the angle (including pitch angle and azimuth angle) of the vehicle target from the data set based on the DOA detection algorithm, and transmits the obtained angle information to the target tracking unit and the information output unit; the target tracking unit uses the Kalman filter algorithm and the track initiation algorithm to track multiple vehicle targets based on the vehicle's speed, distance and angle and generates trajectory point data, that is, the vehicle trajectory is obtained, and input into the information output unit; the environment perception unit obtains road environment information from the data set based on the point cloud imaging algorithm, and constructs a three-dimensional image of the road environment, that is, the road environment point cloud image; the information output unit outputs the vehicle's speed, distance, angle, trajectory and road environment point cloud image to the cloud server unit.

[0060] Reference Figure 5 ,The cloud server unit mainly includes a cloud server, in which all road data sent by the 4D millimeter wave radar ,system unit is stored, including road vehicle information, vehicle trajectory data, ,road environment point cloud images, data obtained after processing by the road ,traffic information service unit, etc., and the road traffic information ,service unit that processes the above information is run ed as a platform for processing ,road data.

[0061] Reference Figure 6The road traffic information service unit of the present invention determines traffic accidents, extracts road features, and identifies abnormal conditions and dangerous driving vehicles on the road based on the road data stored in the cloud server. It mainly includes: a traffic information acquisition unit, a traffic accident determination unit, a dangerous driving detection unit, a road anomaly detection unit, a road feature extraction unit, and a data integration and transmission unit. The traffic information acquisition unit obtains the aforementioned road data to generate road traffic service information. The traffic accident determination unit determines whether an accident has occurred on the road; the road feature extraction unit obtains the traffic volume, queue length, and average speed on the road as road features; the road anomaly detection unit detects abnormal conditions such as obstacles, falling rocks, collapses, and damage on the road; and the dangerous driving detection unit detects target vehicles involved in dangerous driving and their specific violation data. Finally, the vehicle information obtained from the radar and the results obtained from the traffic accident determination unit, road feature extraction unit, road anomaly detection unit, and dangerous driving detection unit are transmitted to the data integration and transmission unit. All traffic road information is uploaded to the relevant departments for traffic management, and traffic road information that can assist driving is sent to the user terminal unit installed on the vehicle.

[0062] Specifically, the traffic accident identification unit of the present invention uses a traffic accident identification and classification algorithm based on a generalized regression neural network to process a road environment point cloud image to identify a traffic accident, label the area where the traffic accident occurred, and transmit the traffic accident information to a data integration and transmission unit. The traffic accident information includes at least a road environment point cloud image labeled with the traffic accident area. The algorithm includes:

[0063] Firstly, the spatial first-order and second-order partial derivatives of the acquired road environment point cloud image are extracted as features for traffic accident discrimination and classification. The features are used to determine the input nodes of the generalized regression neural network and initialize the neural network.

[0064] Secondly, images containing various types of traffic accidents are selected from the stored road environment point cloud image library to form training samples for road anomaly detection. The generalized regression neural network is trained and the actual output of the network and the state of neurons in each layer are determined.

[0065] Then, the gradient descent algorithm is used to determine the weights and thresholds between the hidden layer and the output layer of the generalized regression neural network. The weights and thresholds are updated according to the output error to correct the nonlinear transfer function inside the hidden layer nodes of the generalized neural network.

[0066] Finally, the error of traffic accident discrimination and classification is calculated. If the traffic accident discrimination and classification error meets the actual application requirements, the generalized regression neural network terminates learning, otherwise the image is reselected for learning. At this point, the construction of the traffic accident discrimination classifier is completed. Traffic accidents are discriminated based on the constructed traffic accident discrimination classifier. The dangerous driving behaviors it identifies include speeding, forced lane changes, forced overtaking, and driving on the wrong side of the road.

[0067] The dangerous driving detection unit of the present invention matches the vehicle trajectory with the road environment point cloud image in time and space to generate a complete traffic image, and uses a deep learning algorithm based on a generalized regression neural network to process the matched traffic image. By establishing a dangerous driving recognition model, the target vehicles involved in dangerous driving and their specific violations (including speeding, forced lane changes, forced overtaking, driving on the wrong side of the road, etc.) are labeled, and the dangerous driving data is transmitted to the data integration and sending unit.

[0068] The road anomaly detection unit of the present invention uses an image detection algorithm based on YOLOv3 to process road environment point cloud images, trains a program to identify abnormal conditions in the road environment point cloud images, labels the areas where the road anomalies are located and their possible abnormal conditions, and transmits the road anomaly data to the data integration and transmission unit. The road anomalies identified include obstacles, fallen rocks, collapses, damage, etc. The present invention realizes the identification of road anomaly data by using road environment point cloud images that mark abnormal conditions. The method is as follows:

[0069] The stored road environment point cloud image is used as a data set, and features are extracted from the data set through a basic feature extraction network to obtain a feature map;

[0070] A grid of the same size as the feature map is set on the input image. If the center coordinates of the target to be detected, i.e., the road anomaly, fall within this grid, it will be predicted by this grid.

[0071] Three associated boxes are predicted for each grid, and the associated box most relevant to the target to be detected is finally selected to mark the target.

[0072] The road feature extraction unit of the present invention uses the vehicle speed, distance and vehicle trajectory to obtain the traffic flow, queue length and average road speed on the road as road features through data analysis, and transmits the road feature data to the data integration and transmission unit.

[0073] The data integration and sending unit of the present invention summarizes traffic accident data, dangerous driving data, road anomaly data, road feature data, as well as road vehicle information, vehicle trajectory information and road environment data by time, stores and uploads them to the traffic management department for query and download at any time, and transmits dangerous driving data, road anomaly data, vehicle trajectory information, and road environment data to user terminal units within the service range to provide useful road information for drivers and automatic driving systems.

[0074] Reference Figure 7 The user terminal unit of the present invention receives traffic road information data (including dangerous driving data on any road section, road anomaly data, vehicle trajectory information, road environment point cloud images, and traffic accident data) transmitted from a cloud server, converts it into a 3D model and displays it to the driver, providing auxiliary information for the autonomous driving system. It mainly includes: an information receiving unit, a 3D imaging unit, and an autonomous driving assistance unit.

[0075] The information receiving unit receives information processed by the road traffic information service unit, including dangerous driving data, road anomaly data, vehicle trajectory information, road environment point cloud images, and traffic accident data for any road section, and transmits this information to the 3D imaging unit and the automated driving assistance unit. The 3D imaging unit matches, calculates, and fits the received traffic accident data, dangerous driving data, road anomaly data, vehicle trajectory information, and road environment data to create a 3D image model. Ultimately, it transmits the model of the vehicle on the road, the road environment model, and the dangerous driving warning model to the vehicle terminal for display and to the automated driving assistance unit. The automated driving assistance unit then processes the received model data into a data format that can be processed by the automated driving system and transmits it to the automated driving system via the terminal. The data is then combined with the information independently acquired by the automated driving system to assist the automated driving system.

[0076] In summary, the present invention can provide comprehensive road information to relevant departments for collaborative road management, and can provide assisted driving data to drivers and autonomous driving systems in real time to improve driving safety.

Claims

1. A vehicle-road cooperative system based on 4D millimeter-wave radar, characterized in that: include: 4D millimeter wave radar system unit, used to obtain vehicle information and road environment information and generate road data; the vehicle information includes the speed, distance, angle and height of the vehicle on the road; the road environment information includes the distance, angle and height information obtained by the radar for various static targets on the road; the road data includes vehicle information, vehicle trajectory and road environment point cloud image; The cloud server unit receives and stores all road data sent by the 4D millimeter wave radar system unit and serves as a platform for processing road data; The road traffic information service unit determines traffic accidents, extracts road features, and identifies abnormal conditions and dangerous driving vehicles on the road based on the road data stored in the cloud server to obtain traffic road information data; the road features include vehicle flow, queue length, and average speed; the traffic road information data includes dangerous driving data of any road section, road abnormality data, vehicle trajectory information, road environment point cloud images, and traffic accident data; The user terminal unit receives traffic road information data transmitted by the cloud server, converts it into a 3D model to display to the driver, and provides auxiliary information for the autonomous driving system; Wherein, the road traffic information service unit includes: A traffic information acquisition unit, which acquires the road data; The traffic accident identification unit processes the road environment point cloud image by using the traffic accident identification classification algorithm based on the generalized regression neural network to realize traffic accident identification, labels the area where the traffic accident is located, and transmits the traffic accident information to the data integration and transmission unit; the traffic accident information includes: the road environment point cloud image with the traffic accident area range labeled; The dangerous driving detection unit matches the vehicle trajectory with the road environment point cloud image in time and space to generate a complete traffic image, processes the matched traffic image using a deep learning algorithm based on a generalized regression neural network, labels the target vehicles involved in dangerous driving and their specific violations by establishing a dangerous driving recognition model, and transmits the dangerous driving data to the data integration and transmission unit; The road anomaly detection unit processes the road environment point cloud image using the image detection algorithm based on YOLOv3, trains the program to identify abnormal conditions in the road environment point cloud image, labels the road anomaly area and its possible abnormal conditions, and transmits the road anomaly data to the data integration and transmission unit; The road feature extraction unit uses the speed, distance and trajectory of the vehicle to obtain the traffic volume, queue length and average speed on the road as road features through data analysis, and transmits the road feature data to the data integration and transmission unit; Data integration and transmission unit: summarizes traffic accident data, dangerous driving data, road abnormality data, road feature data, as well as road vehicle information, vehicle trajectory information and road environment data by time, stores and uploads them to the traffic management department for query at any time, and transmits dangerous driving data, road abnormality data, vehicle trajectory information, road environment data to the user terminal unit to provide useful road information for drivers and autonomous driving systems; The traffic accident identification and classification algorithm includes: Firstly, the spatial first-order and second-order partial derivatives of the acquired road environment point cloud image are extracted as the features for traffic accident discrimination and classification. The input nodes of the generalized regression neural network are determined by using the features to initialize the neural network. Secondly, images containing various types of traffic accidents are selected from the stored road environment point cloud image library to form training samples for road anomaly detection to learn the generalized regression neural network, and determine the actual output of the network and the state of neurons in each layer; Then, the weights and thresholds between the hidden layer and the output layer of the generalized regression neural network are determined by using the gradient descent algorithm, and the weights and thresholds are updated according to the output error to correct the nonlinear transfer function inside the hidden layer nodes of the generalized neural network. Finally, the error of traffic accident identification and classification is calculated. If the error of traffic accident identification and classification reaches the requirements of practical application, the generalized regression neural network terminates learning, otherwise the image is reselected for learning. At this point, the construction of the traffic accident identification classifier is completed, and traffic accidents are identified based on the constructed traffic accident identification classifier.

2. The vehicle-road cooperative system based on 4D millimeter-wave radar according to claim 1 is characterized in that: The 4D millimeter wave radar system unit includes: Antenna unit, which transmits electromagnetic waves and receives target echoes to obtain raw data of the vehicle and road environment; A signal processing unit, processing the raw data to obtain a data set including vehicle information and road environment information; A speed and distance extraction unit, based on a CFAR detection algorithm, obtains the speed and distance of the vehicle from the data set; An angle calculation unit, based on a DOA detection algorithm, obtains the angle of the vehicle from the data set; the angle of the vehicle includes its pitch angle and azimuth angle; The target tracking unit uses the Kalman filter algorithm and the track initiation algorithm to track multiple targets and generate vehicle trajectories based on the vehicle's speed, distance and angle; The environmental perception unit processes the road environment information in the data set based on the point cloud imaging algorithm and constructs a three-dimensional image of the road environment, namely, a road environment point cloud image; The information output unit outputs the vehicle's speed, distance, angle, vehicle trajectory and road environment point cloud image to the cloud server unit.

3. The vehicle-road cooperative system based on 4D millimeter-wave radar according to claim 1 is characterized in that: The dangerous driving behaviors that can be identified by the dangerous driving identification model include speeding, forced lane change, forced overtaking, and driving on the wrong side of the road; the road anomalies include obstacles, falling rocks, collapse, and damage.

4. The vehicle-road cooperative system based on 4D millimeter-wave radar according to claim 1 is characterized in that: The recognition of the road abnormality data is to mark the road environment point cloud image of the abnormal condition, and the method is as follows: The stored road environment point cloud image is used as a data set, and features are extracted from the data set through a basic feature extraction network to obtain a feature map; A grid of the same size as the feature map is set on the input image. If the center coordinates of the target to be detected, i.e., the abnormal road condition, fall within the grid, the grid will be used to predict it. Three associated boxes are predicted for each grid, and finally the associated box most relevant to the target to be detected is selected to mark the target.

5. The vehicle-road cooperative system based on 4D millimeter-wave radar according to claim 1 is characterized in that: The user terminal unit comprises: Information receiving unit: receiving processed information from the road traffic information service unit, including: dangerous driving data of any road section, road abnormality data, vehicle trajectory information, road environment point cloud images, and traffic accident data; 3D imaging unit: matches, calculates and fits the received traffic accident data, dangerous driving data, road abnormality data, vehicle trajectory information, and road environment data to establish a 3D image model, and finally transmits the driving vehicle model, road environment model, and dangerous driving warning model to the vehicle terminal for display and to the automatic driving assistance unit; Autonomous driving assistance unit: processes the received model data into a data format that can be processed by the autonomous driving system, transmits it to the autonomous driving system through the terminal, and combines it with the information autonomously acquired by the autonomous driving system to assist the autonomous driving system.

Images