Vehicle real-time monitoring and early warning method and system based on Beidou high precision

Abstract

The invention relates to the technical field of intelligent traffic, in particular to a vehicle real-time monitoring and early warning method and system based on Beidou high precision, and the method comprises the steps: collecting Beidou dual-frequency original observation data, vehicle CAN bus data and inertial measurement data; calculating the centimeter-level high-precision pose state of the vehicle by adopting a tight coupling fusion algorithm based on factor graph optimization; determining a lane level position through high-precision map matching, and combining real-time traffic and environment information to form a dynamic context; extracting multi-dimensional driving behavior characteristics and risk indexes; carrying out comprehensive risk assessment by using a deep learning model mixed with an attention mechanism and a long-short-term memory network, and outputting a risk level; and finally, according to the risk level and the dynamic context, graded early warning information is generated in a self-adaptive manner, and the graded early warning information is collaboratively published to a vehicle-mounted terminal, a cloud control platform and surrounding vehicles through multiple channels.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, specifically to a method and system for real-time vehicle monitoring and early warning based on BeiDou high precision. Background Technology

[0002] With the surge in car ownership and the increasing complexity of road traffic, traffic accidents have become a global public safety issue. Traditional vehicle safety systems, such as anti-lock braking systems (ABS) and electronic stability programs (ESCs), are passive response controls that cannot intervene before a danger occurs. Existing vehicle monitoring and early warning systems mainly suffer from the following technical bottlenecks: First, the perception accuracy is insufficient. The widely used single-point GPS positioning technology has an accuracy of only a few meters to ten meters, which is insufficient to distinguish the vehicle's lane position, resulting in low reliability of location-based warnings (such as lane departure warnings). Second, the risk assessment dimensions are limited. Most systems rely on single threshold alarms such as speeding and driver fatigue, lacking the fusion analysis and comprehensive judgment of multi-dimensional information such as vehicle dynamics, driver behavior patterns, real-time traffic environment, and road geometry. The warning models are simplistic, leading to frequent false alarms and missed alarms. Third, the warning methods are rigid. Warning information lacks tiered and context-adaptive features, typically released in a uniform audio-visual format. It cannot provide differentiated and personalized prompts based on risk level, driver status, and external environment, potentially causing drivers to experience "alarm fatigue" from frequent false alarms and thus ignore genuine high-risk warnings.

[0003] As one of the world's four major satellite navigation systems, the BeiDou Navigation Satellite System provides real-time high-precision positioning capabilities at the centimeter to decimeter level through its BeiDou Precise Point Positioning Service and BeiDou Ground-Based Augmentation System. This provides a new technological foundation for overcoming the aforementioned bottlenecks in perception accuracy. However, high-precision location information is only one of the basic inputs for building the next generation of intelligent early warning systems. How to deeply and effectively integrate this high-precision spatiotemporal information with vehicle bus data, inertial sensor data, high-precision dynamic maps, and real-time traffic information, and on this basis, build a complete technological system capable of understanding complex driving scenarios, predicting potential collision risks, and generating intelligent decision-making and early warning systems, remains a critical technical challenge to be overcome in this field. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for real-time vehicle monitoring and early warning based on BeiDou high precision, so as to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a real-time vehicle monitoring and early warning method based on BeiDou high precision, comprising the following steps: S1. Multi-source heterogeneous data acquisition: Real-time acquisition of raw observation data from the BeiDou Navigation Satellite System, vehicle controller local area network bus data, and inertial measurement unit data from the vehicle end; wherein, the raw observation data from the BeiDou Navigation Satellite System includes at least pseudorange and carrier phase observation values ​​of L-band and B2a dual frequencies; S2. High-precision state perception and data fusion: A tightly coupled fusion algorithm based on factor graph optimization is adopted to fuse the multi-source heterogeneous data in step S1, and to calculate in real time the vehicle's position, three-dimensional velocity, three-dimensional attitude angle and vehicle motion state information under non-holonomic constraints in three-dimensional space with centimeter-level precision. S3. High-precision map matching and dynamic context awareness: Using the high-precision location information output in step S2, it is matched with a high-precision vector electronic map containing lane-level geometric topology and traffic sign semantic information to determine the precise lane where the vehicle is located; at the same time, real-time traffic event information, lane-level traffic flow parameters and micro-meteorological environment information are integrated to form the dynamic context environment for vehicle driving. S4. Real-time extraction of multi-dimensional risk features: Based on the high-frequency time-series state information obtained in step S2, micro-driving behavior feature vectors reflecting the longitudinal, lateral and vertical motion of the vehicle are extracted through a preset sliding time window; and based on the dynamic context environment obtained in step S3, diversified risk indicators including collision time, vehicle headway, lane keeping degree, traffic rule violation degree and environmental visibility impact coefficient are calculated in real time. S5. Comprehensive risk assessment based on machine learning: Construct a deep neural network model containing a temporal feature input layer. After concatenating and normalizing the driving behavior feature vector extracted in step S4 with diversified risk indicators, input the model into the deep neural network model. The model outputs a continuous risk value that represents the comprehensive risk level of the vehicle at the current moment, and maps the risk value to a discrete risk level according to a preset threshold. S6. Context-Adaptive Hierarchical Early Warning Decision-Making and Issuance: Based on the risk level and risk type determined in step S5 and the dynamic context in step S3, early warning information containing specific risk descriptions, avoidance suggestions, and expected consequences is generated from a preset multi-level early warning strategy library; through multiple channels of concurrent distribution via vehicle bus system, cellular network, and vehicle wireless communication technology, the early warning information is adaptively distributed to the in-vehicle human-machine interface, cloud monitoring platform, and in-vehicle terminals of relevant traffic participants in the surrounding area.

[0006] Furthermore, the tightly coupled fusion algorithm based on factor graph optimization described in step S2 is as follows: using the error state of the inertial measurement unit as the graph optimization factor, and simultaneously using the geometric distance factor constructed from the Beidou dual-frequency observations, the Doppler frequency shift factor, and the wheel speed factor provided by the vehicle CAN bus as constraints, a global factor graph is constructed, and the optimal state estimate of the vehicle is solved in real time through sliding window optimization.

[0007] Furthermore, the extraction of micro-driving behavior feature vectors in step S4 specifically includes: extracting acceleration jerk and following distance variation coefficient in the longitudinal dimension; extracting lane departure cumulative time and steering wheel angle entropy in the lateral dimension; and extracting the vehicle vertical acceleration spectrum features caused by road surface unevenness excitation in the vertical dimension.

[0008] Furthermore, the deep neural network model mentioned in step S5 is a hybrid model of long short-term memory network and attention mechanism; the long short-term memory network is used to learn the temporal dependency relationship between driving behavior and risk indicators, and the attention mechanism is used to dynamically weight the contribution of different time steps and different feature dimensions to the current risk assessment.

[0009] Furthermore, the construction of the multi-level early warning strategy library in step S6 is based on historical accident data and expert experience, and uses the decision tree method to perform rule-based mapping and storage of different risk scenarios, risk levels, optimal early warning content, release channels and release intensity.

[0010] Secondly, the present invention provides a BeiDou-based high-precision vehicle real-time monitoring and early warning system for implementing the above method, comprising: The vehicle-mounted multi-source sensing module, used to execute step S1, includes a Beidou high-precision positioning unit, a vehicle bus data acquisition unit, and an inertial measurement unit; The edge computing fusion module is used to execute step S2. It is deployed on the vehicle-mounted edge computing platform and has a built-in tightly coupled fusion algorithm based on factor graph optimization. The high-precision dynamic map service module is used to execute step S3. It is deployed in the cloud and provides a fusion query interface for high-precision map data, real-time traffic information and meteorological information. The cloud-based risk analysis engine module, used to execute steps S4 and S5, is deployed in a cloud computing center and includes a feature extraction submodule and a deep neural network risk assessment submodule. The early warning decision and collaborative distribution module, used to execute step S6, is deployed in the cloud or at the edge and includes an early warning strategy library management submodule and a multi-channel communication gateway.

[0011] Furthermore, the BeiDou high-precision positioning unit in the vehicle-mounted multi-source sensing module integrates a radio frequency baseband integrated chip that supports the full signal mode of the BeiDou-3 system, and can receive and decode the precise single-point positioning service signal broadcast by BeiDou satellites in real time.

[0012] Furthermore, the multi-channel communication gateway in the early warning decision and collaborative distribution module supports cloud communication based on cellular networks, direct communication based on vehicle wireless communication technology, and local broadcast communication based on dedicated short-range communication, and can dynamically select the communication link according to the urgency of the early warning and the receiving target.

[0013] Compared with the prior art, the beneficial effects achieved by the present invention are: A fundamental leap in perception accuracy has been achieved: by tightly coupling and fusing the factor graph of Beidou dual-frequency observations and inertial data, the accuracy limit of traditional single-point positioning or loose combination filtering has been broken through, realizing stable and reliable centimeter-level continuous lane tracking, laying an indisputable accurate data foundation for all upper-level early warning applications.

[0014] A multi-dimensional risk assessment system has been constructed: abandoning the single threshold judgment, it creatively integrates high-precision motion state, micro driving behavior, high-precision static map semantics and dynamic traffic environment four-dimensional information, and adopts advanced attention mechanism LSTM network for modeling, so that risk assessment can understand scene context and remember behavior patterns, realizing the leap from "perception" to "cognition", and significantly improving the accuracy and predictability of early warning.

[0015] It achieves highly intelligent early warning decision-making and collaboration: the proposed scenario-adaptive hierarchical early warning mechanism transforms early warning from a simple signal trigger into a comprehensive decision-making process based on risk level, type, and environment. Combined with V2X multi-channel dissemination capabilities, it not only enables personalized warnings within vehicles but also extends to collaborative early warnings between vehicles and between vehicles and roads, constructing a safety protection network for local traffic environments and greatly enhancing the system's proactive protection effectiveness.

[0016] The system boasts an advanced architecture that combines high performance and scalability: the cloud-edge-device collaborative architecture deploys high-load fusion positioning and model inference at the edge or cloud, ensuring the operation of complex algorithms; at the same time, the modular design allows high-precision map services, risk models, and other components to be upgraded and iterated independently, giving the system excellent evolution capabilities and adaptability to different vehicle models and scenarios. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1This is a complete flowchart of the vehicle real-time monitoring and early warning method provided in the embodiments of the present invention; Figure 2 This is a cloud-edge-device collaborative architecture block diagram of the vehicle real-time monitoring and early warning system provided in an embodiment of the present invention.

[0018] In the diagram: 310, Vehicle-mounted terminal; 311, High-precision BeiDou positioning module; 312, Vehicle network gateway; 313, IMU; 320, Edge computing node; 321, High-precision fusion positioning engine; 330, Cloud platform; 331, High-precision dynamic map service; 332, AI risk analysis engine; 333, Early warning decision center; 334, Collaborative communication gateway; 340, Execution and presentation terminal; 341, Human-computer interaction interface; 324, V2X communication equipment; 343, Cloud monitoring screen. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] This invention provides a technical solution: a real-time vehicle monitoring and early warning method based on BeiDou high precision, comprising the following six core technical steps, forming a closed loop from perception to decision execution: S1. Multi-source heterogeneous data acquisition: The system synchronously acquires three types of core data streams at the vehicle end. The first stream is raw observation data (pseudorange, carrier phase, and Doppler) from the dual-frequency Beidou receiver, which is the foundation of high-precision positioning. The second stream is vehicle dynamic parameters read through the controller's local area network interface, such as vehicle speed, wheel speed, steering wheel angle, and yaw rate. The third stream is raw triaxial acceleration and angular velocity data from the miniature inertial measurement unit. All data requires time synchronization and calibration.

[0021] S2. High-Precision State Perception and Data Fusion: This step is crucial for improving perception accuracy. A tightly coupled fusion architecture based on factor graph optimization is adopted to unify the modeling of heterogeneous data from S1. The error state of the inertial navigation system is used as the main body of graph optimization. The geometric distance factor and Doppler velocity factor constructed from BeiDou observations are used as absolute measurement constraints, the vehicle wheel speed factor is used as a relative constraint, and the non-holonomic motion constraint factor of the vehicle is introduced. The sliding window optimizer solves in real time, outputting a six-degree-of-freedom state estimate of the vehicle with a frequency higher than 100Hz, including centimeter-level 3D position, millimeter / second-level 3D velocity, and 3D attitude angles better than 0.1 degrees.

[0022] S3. High-Precision Map Matching and Dynamic Context Awareness: This section matches the high-precision location output from S2 with a pre-loaded or online streaming high-precision vector map. This map must have lane-level accuracy, including lane geometry, curvature, slope, and the 3D location and semantics of traffic signs. The matching algorithm employs a sequence matching method based on a Hidden Markov Model to determine the precise lane ID of the vehicle. Simultaneously, the system accesses real-time dynamic information released by the traffic management center via V2X communication or mobile networks, such as information on upcoming accidents, construction, and congestion, as well as local weather information released by the meteorological department, forming a comprehensive understanding of the current driving environment.

[0023] S4. Real-time Extraction of Multi-Dimensional Risk Features: Based on the high-frequency, high-precision state sequence output from S2, feature engineering is performed within a sliding time window. Vertical risk features include, but are not limited to: collision time and headway calculated based on relative distance and speed, and their rate of change; acceleration abruptness reflecting the degree of driving aggression. Lateral risk features include: lateral offset distance and deviation speed calculated based on lane line equations; statistical entropy values ​​of the steering wheel angle sequence, used to quantify the smoothness of steering operations. Combined with the context of S3, rule compliance features are calculated, such as the difference between the current vehicle speed and the speed limit and curvature-recommended speed in the map.

[0024] S5. Comprehensive Risk Assessment Based on Machine Learning: This step is the core of achieving intelligent judgment. A deep learning model specifically designed for driving risk assessment is developed. This model uses a Long Short-Term Memory (LSTM) network as its backbone to capture the temporal evolution patterns of driving behavior characteristics; an attention mechanism layer is introduced at the front end, enabling the model to dynamically focus on the historical moments and feature dimensions that contribute the most to the current risk. The model takes the multi-dimensional feature vector extracted in S4 as input, undergoes multiple nonlinear transformations, and finally outputs a continuous risk probability value between 0 and 1. The system classifies this value into "low risk," "medium risk," and "high risk" levels according to preset risk probability thresholds (e.g., 0.3, 0.6, 0.9).

[0025] S6. Context-Adaptive Tiered Warning Decision-Making and Issuance: The system has a built-in, updatable warning strategy knowledge base. After S5 determines the risk level, the warning decision engine combines the risk type (rear-end collision, lane departure, speeding, etc.) and the dynamic context in S3 (such as weather and road conditions) to retrieve the optimal warning solution from the strategy base. This solution specifies the text / voice template for the warning content, the issuance channel (in-vehicle audio, display screen, seat vibration, V2X broadcast, cloud push), and the issuance intensity. For example, for a high-risk lane departure on a highway, the system may simultaneously trigger strong steering wheel vibration, a voice warning, and broadcast collaborative warning information to vehicles in adjacent lanes.

[0026] A BeiDou-based high-precision vehicle real-time monitoring and early warning system for implementing the above method includes: The vehicle-mounted multi-source sensing module is used to execute step S1. It includes a Beidou high-precision positioning unit, a vehicle bus data acquisition unit, and an inertial measurement unit. The Beidou high-precision positioning unit integrates a radio frequency baseband integrated chip that supports the full signal mode of the Beidou-3 system. It can receive and decode the precise single-point positioning service signal broadcast by Beidou satellite in real time. The edge computing fusion module is used to execute step S2. It is deployed on the vehicle-mounted edge computing platform and has a built-in tightly coupled fusion algorithm based on factor graph optimization. The high-precision dynamic map service module is used to execute step S3. It is deployed in the cloud and provides a fusion query interface for high-precision map data, real-time traffic information and meteorological information. The cloud-based risk analysis engine module, used to execute steps S4 and S5, is deployed in a cloud computing center and includes a feature extraction submodule and a deep neural network risk assessment submodule. The early warning decision and collaborative distribution module, used to execute step S6, is deployed in the cloud or at the edge. It includes an early warning strategy library management submodule and a multi-channel communication gateway. The multi-channel communication gateway supports cloud communication based on cellular networks, direct communication based on vehicle wireless communication technology, and local broadcast communication based on dedicated short-range communication. It can dynamically select the communication link according to the urgency of the early warning and the receiving target.

[0027] Example 1: Application of a BeiDou-based high-precision vehicle real-time monitoring and early warning method in a highway scenario See Figure 1 This embodiment demonstrates how the system works when a passenger vehicle is driving on a highway in the rain.

[0028] S1: The vehicle terminal collects Beidou B1I / B2a dual-frequency observations at a frequency of 100Hz, vehicle speed (115km / h) on the CAN bus, yaw rate provided by the ESP system, and raw acceleration and angular velocity data from the IMU.

[0029] S2: Edge computing unit operation factor graph optimization algorithm. The algorithm uses IMU data for inertial estimation, while effectively suppressing ionospheric delay errors using BeiDou dual-frequency observations, and constraining inertial drift using wheel speed information. After optimization, the algorithm outputs the vehicle's precise position at this moment: 0.2 meters to the right of the center line of the second lane southbound on the G4 Beijing-Hong Kong-Macau Expressway; precise speed: 114.8 km / h; heading angle: 182.3 degrees; and detects continuous negative longitudinal acceleration (slight deceleration).

[0030] S3: The high-precision map service confirms that the road section has a gentle curvature, but the current weather is moderate rain, and the "recommended speed limit for slippery roads" associated with the map is 100km / h. V2X communication received a real-time event that a disabled vehicle is stopped in the first lane 800 meters ahead.

[0031] S4: The risk feature extraction module analyzes the data from the most recent 5 seconds. Longitudinal features: The calculated TTC (Traffic Time Between) between this vehicle and the vehicle in the same lane ahead is 6.5 seconds, but it is decreasing; the acceleration abruptness is high, indicating a sudden change in driving operation. Lateral features: The vehicle position continues to drift slightly to the right of the lane centerline, and the deviation index is increasing. Contextual features: The current speed (114.8 km / h) is significantly higher than the recommended speed limit for rainy weather (100 km / h).

[0032] S5: Input the above feature vectors into the trained Attention-LSTM model. The model's attention layer shows that it assigns high weights to "speeding coefficient" and "deviation". The model's overall judgment: Under rainy conditions with poor visibility and slippery roads, speeding accompanied by unstable lane keeping, and with potential obstacles ahead, the risk probability is calculated to be 0.78.

[0033] S6: The warning strategy library generates the highest-level warning strategy based on the combination of conditions such as "high risk (0.78)," "rainy weather," "highway," and "speeding & lane departure." The system executes: 1) Displays "Danger! Please slow down immediately and stay in your lane!" in flashing red on the HUD; 2) Announces the warning content in an urgent tone via voice synthesis; 3) Triggers a short and forceful steering wheel vibration; 4) Broadcasts the vehicle's precise location, status, and high-risk warning via V2X for following vehicles to receive.

[0034] Example 2: Application of the system in urban intersection scenarios As a bus approaches an intersection where the traffic light is about to turn red, system S2 calculates that its speed has not decreased sufficiently. S3 map matching confirms that crossing the line at a red light is prohibited. S4 feature extraction shows a surge in the "red light running tendency" indicator. S5 model assessment indicates a rapid increase in risk. S6 the system issues a strong audible and visual warning "Red light!" in the bus driver's cab and sends a "conflict warning" message via V2X to other vehicles about to cross the intersection with a green light, prompting them to be cautious.

[0035] Example 3: A Real-time Vehicle Monitoring and Early Warning System Based on BeiDou High Precision See Figure 2 The system adopts a three-layer architecture of cloud-edge-device: Vehicle-mounted terminal 310: includes a high-precision Beidou positioning module 311, a vehicle network gateway 312, and an IMU 313, which are responsible for raw data acquisition and initial packaging.

[0036] Edge computing node 320: Deployed on roadside units or vehicle-mounted high-performance computing platforms. It contains a high-precision fusion positioning engine 321, which runs the factor graph optimization algorithm to provide localized high-precision status output.

[0037] Cloud Platform 330: This is the system's brain. It includes: High-precision dynamic map service 331, providing global map and real-time information fusion; AI risk analysis engine 332, which undertakes feature extraction and deep learning model inference tasks; Early warning decision center 333, which manages the policy library and generates final instructions; and collaborative communication gateway 334, which is responsible for communication with vehicles, other cloud platforms, and traffic management centers.

[0038] The warning information is ultimately applied to the execution and presentation end 340, including the human-computer interaction interface 341, V2X communication equipment 342, and cloud monitoring screen 343.

[0039] Through collaborative work, the system achieves a reasonable allocation of complex data processing in the cloud, real-time control decision-making at the edge, and reliable perception at the vehicle end, ensuring the overall high performance, low latency, and high reliability of the system.

[0040] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for real-time vehicle monitoring and early warning based on BeiDou high-precision positioning, characterized in that, Includes the following steps: S1. Multi-source heterogeneous data acquisition: Real-time acquisition of raw observation data from the BeiDou Navigation Satellite System, vehicle controller local area network bus data, and inertial measurement unit data from the vehicle end; wherein, the raw observation data from the BeiDou Navigation Satellite System includes at least pseudorange and carrier phase observation values ​​of L-band and B2a dual frequencies; S2. High-precision state perception and data fusion: A tightly coupled fusion algorithm based on factor graph optimization is adopted to fuse the multi-source heterogeneous data in step S1, and to calculate in real time the vehicle's position, three-dimensional velocity, three-dimensional attitude angle and vehicle motion state information under non-holonomic constraints in three-dimensional space with centimeter-level precision. S3. High-precision map matching and dynamic context awareness: Using the high-precision location information output in step S2, it is matched with a high-precision vector electronic map containing lane-level geometric topology and traffic sign semantic information to determine the precise lane where the vehicle is located; at the same time, real-time traffic event information, lane-level traffic flow parameters and micro-meteorological environment information are integrated to form the dynamic context environment for vehicle driving. S4. Real-time extraction of multi-dimensional risk features: Based on the high-frequency time-series state information obtained in step S2, micro-driving behavior feature vectors reflecting the longitudinal, lateral and vertical motion of the vehicle are extracted through a preset sliding time window; and based on the dynamic context environment obtained in step S3, diversified risk indicators including collision time, vehicle headway, lane keeping degree, traffic rule violation degree and environmental visibility impact coefficient are calculated in real time. S5. Comprehensive risk assessment based on machine learning: Construct a deep neural network model containing a temporal feature input layer. After concatenating and normalizing the driving behavior feature vector extracted in step S4 with diversified risk indicators, input the model into the deep neural network model. The model outputs a continuous risk value that represents the comprehensive risk level of the vehicle at the current moment, and maps the risk value to a discrete risk level according to a preset threshold. S6. Context-Adaptive Hierarchical Early Warning Decision-Making and Issuance: Based on the risk level and risk type determined in step S5 and the dynamic context in step S3, early warning information containing specific risk descriptions, avoidance suggestions, and expected consequences is generated from a preset multi-level early warning strategy library; through multiple channels of concurrent distribution via vehicle bus system, cellular network, and vehicle wireless communication technology, the early warning information is adaptively distributed to the in-vehicle human-machine interface, cloud monitoring platform, and in-vehicle terminals of relevant traffic participants in the surrounding area.

2. The method for real-time vehicle monitoring and early warning based on BeiDou high precision as described in claim 1, characterized in that, The tightly coupled fusion algorithm based on factor graph optimization described in step S2 is as follows: the error state of the inertial measurement unit is used as the graph optimization factor, and the geometric distance factor constructed from the Beidou dual-frequency observations, the Doppler frequency shift factor, and the wheel speed factor provided by the vehicle CAN bus are used as constraints to construct a global factor graph, and the optimal state estimate of the vehicle is solved in real time through sliding window optimization.

3. The method for real-time vehicle monitoring and early warning based on BeiDou high precision as described in claim 1, characterized in that, Step S4, which involves extracting the micro-driving behavior feature vector, specifically includes: extracting acceleration abruptness and following distance variation coefficient in the longitudinal dimension; extracting lane departure cumulative time and steering wheel angle entropy in the lateral dimension; and extracting the vehicle vertical acceleration spectrum feature caused by road surface unevenness excitation in the vertical dimension.

4. The method for real-time vehicle monitoring and early warning based on BeiDou high precision as described in claim 1, characterized in that, The deep neural network model mentioned in step S5 is a hybrid model of long short-term memory network and attention mechanism; the long short-term memory network is used to learn the temporal dependency relationship between driving behavior and risk indicators, and the attention mechanism is used to dynamically weight the contribution of different time steps and different feature dimensions to the current risk assessment.

5. The method for real-time vehicle monitoring and early warning based on BeiDou high precision as described in claim 1, characterized in that, The construction of the multi-level early warning strategy library in step S6 is based on historical accident data and expert experience. It uses the decision tree method to perform rule-based mapping and storage of different risk scenarios, risk levels, optimal early warning content, release channels, and release intensity.

6. A BeiDou high-precision vehicle real-time monitoring and early warning system for implementing the method of any one of claims 1 to 5, characterized in that, include: The vehicle-mounted multi-source sensing module, used to execute step S1, includes a Beidou high-precision positioning unit, a vehicle bus data acquisition unit, and an inertial measurement unit; The edge computing fusion module is used to execute step S2. It is deployed on the vehicle-mounted edge computing platform and has a built-in tightly coupled fusion algorithm based on factor graph optimization. The high-precision dynamic map service module is used to execute step S3. It is deployed in the cloud and provides a fusion query interface for high-precision map data, real-time traffic information and meteorological information. The cloud-based risk analysis engine module, used to execute steps S4 and S5, is deployed in a cloud computing center and includes a feature extraction submodule and a deep neural network risk assessment submodule. The early warning decision and collaborative distribution module, used to execute step S6, is deployed in the cloud or at the edge and includes an early warning strategy library management submodule and a multi-channel communication gateway.

7. A real-time vehicle monitoring and early warning system based on BeiDou high precision as described in claim 6, characterized in that, The BeiDou high-precision positioning unit in the vehicle-mounted multi-source sensing module integrates a radio frequency baseband integrated chip that supports the full signal mode of the BeiDou-3 system, and can receive and decode the precise single-point positioning service signal broadcast by BeiDou satellites in real time.

8. A real-time vehicle monitoring and early warning system based on BeiDou high precision as described in claim 6, characterized in that, The multi-channel communication gateway in the early warning decision and collaborative distribution module supports cloud communication based on cellular networks, direct communication based on vehicle wireless communication technology, and local broadcast communication based on dedicated short-range communication. It can also dynamically select the communication link according to the urgency of the early warning and the receiving target.

Images