Intelligent vehicle fusion sensing and decision-making method and system based on low-orbit satellite communication and high-precision positioning

Abstract

The invention discloses an intelligent vehicle fusion perception and decision-making method and system based on low-orbit satellite communication and high-precision positioning, and the method comprises the steps: firstly, collecting the perception data of an intelligent vehicle through the low-orbit satellite communication and high-precision positioning technology, and achieving the time-space consistency processing through a cloud end, the method includes delay compensation, semantic information matching and multi-observation joint state estimation, improves complex scene perception precision, and reduces dangerous decision-making behaviors and high takeover rate caused by perception defects of an intelligent automobile. Secondly, a behavior guiding strategy is constructed based on track data of a large number of vehicle groups, and a macroscopic decision basis is provided for the intelligent vehicle groups; and through a multi-agent reinforcement learning algorithm, the intelligent vehicle group can carry out self-vehicle optimal decision making, so that efficient traffic flow management and group collaborative optimization are realized. A brand new technical scheme is provided for dynamic traffic flow management in a complex traffic scene, traffic safety and efficiency are improved, and important support is provided for application of an intelligent traffic system and automatic driving.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation systems, and particularly to a method and system for realizing the fusion of perception information and collaborative decision-making of intelligent vehicles in complex traffic scenarios by using low-earth orbit satellite communication and high-precision positioning technologies. Background Art

[0002] With the increasing complexity of urban traffic and the continuous improvement of the degree of intelligence, traditional traffic management systems are facing huge challenges, especially in dynamic traffic environments. The rapid development of intelligent transportation systems (ITS) and autonomous driving technologies has promoted the intelligent upgrade of traffic management. How to improve the efficiency of traffic flow, enhance traffic safety, and optimize traffic decisions has become a research hotspot in the current traffic field. As the core component of the autonomous driving system, intelligent vehicles have great potential in real-time perception, decision-making, and collaboration. However, due to the complex and dynamically changing traffic environment, the perception and decision-making capabilities of intelligent vehicles still face many technical problems.

[0003] The rapid development of low-earth orbit satellite communication technology in recent years has provided an efficient, stable, and high-precision positioning and communication solution for intelligent vehicles. Compared with traditional positioning and navigation systems, low-earth orbit satellite systems have the advantages of low latency, high bandwidth, and high-precision positioning, and can provide more accurate positioning data and real-time communication support for intelligent vehicles. This provides new possibilities for the collaborative decision-making and multi-source information fusion of intelligent vehicle groups. However, how to effectively utilize low-earth orbit satellite communication and positioning technologies to achieve the efficient cooperation of intelligent vehicles in complex traffic scenarios is still an urgent problem to be solved.

[0004] Therefore, how to combine multiple heterogeneous perception information based on low-earth orbit satellite communication and high-precision positioning technologies and achieve the collaborative decision-making of intelligent vehicle groups through intelligent algorithms has become the key technology for improving traffic flow efficiency, optimizing traffic safety, and enhancing the performance of intelligent transportation systems. The present invention proposes a method for the fusion perception and decision-making of intelligent vehicles based on low-earth orbit satellite communication and high-precision positioning, aiming to provide a new systematic perception and decision-making solution for intelligent vehicles in complex traffic scenarios, thereby effectively improving the overall efficiency and safety of intelligent transportation systems. Summary of the Invention

[0005] The present invention aims to provide a method and system for the fusion perception and decision-making of intelligent vehicles based on low-earth orbit satellite communication and high-precision positioning, which can realize the fusion and optimized decision-making of multi-source perception information in complex traffic scenarios and improve the overall traffic passing efficiency. The method mainly includes the following two key steps:

[0006] Fusion of heterogeneous perception information: By integrating the perception data of multi-source heterogeneous intelligent vehicles, a comprehensive understanding of the traffic scenario is achieved.

[0007] Collaborative Decision-making and Game Theory: Based on scenario understanding, optimize the game decisions between intelligent vehicles and non-intelligent vehicles, and at the same time achieve collaborative decision-making within the intelligent vehicle system, and finally build an intelligent traffic management system.

[0008] To achieve the above objectives, the present invention adopts the following technical solutions:

[0009] S1 In the first aspect of the present invention, a method for perception result fusion and optimization based on low-earth orbit satellite communication and high-precision positioning capabilities is provided, including two parts: information collection and cloud information processing:

[0010] S11 The information collection part is used to realize the collection and transmission of environmental perception data of heterogeneous intelligent vehicles. In complex traffic scenarios, a communication link between the cloud and each intelligent vehicle is constructed using low-earth orbit satellites. For the heterogeneous differences in sensor configurations and algorithms of different intelligent vehicles, the present invention will adopt a post-fusion method to integrate perception information.

[0011] The information collection content includes the following two parts, including perception data collection, data encoding and transmission:

[0012] The perception data collection includes, but is not limited to, perception data of heterogeneous intelligent vehicles (such as adjacent vehicle states, three-dimensional detection frames, semantic information, etc.); high-precision vehicle state information (such as position, heading angle, speed, acceleration, etc.) calculated through multiple groups of low-earth orbit satellite propagation signals and ground monitoring station signals;

[0013] The data encoding and transmission is to compress and encode the collected data, and then upload it to the cloud server for storage and processing through the communication capabilities of low-earth orbit satellites. The transmission process adopts efficient encoding means (such as entropy coding-based compression algorithms) to reduce bandwidth, while ensuring the real-time and integrity of transmission.

[0014] S12 The cloud information processing part is used to receive the perception state information collected and published by heterogeneous multi-intelligent vehicles in the supervision scenario through low-earth orbit satellite communication in the cloud server, and realize target-level fusion based on high-precision positioning, so as to achieve more accurate scenario understanding in highly uncertain traffic scenarios.

[0015] The main processing strategies of the cloud processing part include time delay compensation, semantic association matching and multi-observation joint state trajectory estimation:

[0016] The time delay compensation is to perform delay compensation for all to-be-estimated targets observed based on the multi-source perception results of heterogeneous intelligent vehicles. The specific steps are as follows:

[0017] Construction of motion prediction model: Construct a motion prediction model for each vehicle based on the motion state (position, speed, acceleration, etc.) of the target object, such as using a uniform motion model or a uniformly accelerated model.

[0018] Calculation of delay time: Calculate the possible motion state delay time Δt of the target object based on data transmission delay and sensing link delay;

[0019] State prediction: Use the motion prediction model to predict the current accurate position of the target object and obtain the compensated state quantity.

[0020] The semantic association matching mentioned above means that in the observation scenario, for multiple intelligent vehicles with sensing capabilities, target matching work is carried out for the overlapping part of their observation ranges. It mainly includes two parts: basic matching based on intersection over union and semantic feature enhanced matching, which are specifically as follows:

[0021] The basic matching based on intersection over union uses the high-precision positioning ability of low-orbit satellites, projects the perceived targets in three-dimensional space according to the spatial positioning of the sensing carrier, and uses the intersection over union matching strategy to achieve basic spatial matching in units of driving scenarios, and constructs a preliminary matching result for heterogeneous intelligent vehicle perception;

[0022] The semantic feature enhanced matching means using high-level semantic information to achieve high-order joint matching. It includes content such as target association feature extraction, matching algorithm design, and dynamic weight adjustment, which are specifically as follows:

[0023] The target association feature extraction combines the positioning information and semantic information of the vehicle, including color, vehicle type, etc., to construct a joint feature vector F = [P, S], where P represents the spatial position feature and S represents the semantic label;

[0024] The matching algorithm design uses a location-based nearest neighbor matching algorithm or a weighted matching method based on semantic information to match the sensing data of each vehicle with the information pre-calibrated in the map. The matching degree calculation formula can be expressed as:

[0025] S 匹配 = αIoU + βS 相似度

[0026] where α and β are weight parameters, IoU is the score of the basic matching based on intersection over union mentioned above, and S 相似度 is the matching similarity score obtained by encoding and decoding the joint feature vector base F based on a neural network:

[0027]

[0028] where f ω is the neural network model, is the semantic joint feature vector collected for N vehicles to be observed;

[0029] For the dynamic weight adjustment, the weights of each information source are adjusted according to environmental conditions (such as light, weather, etc.). For example, when the light is good, the color information has a high credibility and is given a larger weight; in bad weather, the position information is more reliable and the weights are adjusted accordingly.

[0030] For the multi-observation joint state estimation, based on the delay compensation and semantic matching strategies, the present invention introduces a multi-observation joint state estimation strategy, which fuses the perception data of multiple intelligent vehicles and other observation sources to form a unified state estimation of the targets in the traffic scene. The delay compensation unifies the spatio-temporal reference, and the target association algorithm based on semantic information matching optimizes the observation consistency of the state estimation. On this basis, through the Kalman filter or unscented Kalman filter method, all the compensated and matched observation data are fused into the unified state estimation of the target.

[0031] This method improves the accuracy and stability of global perception by integrating multi-dimensional information such as spatio-temporal consistency, semantic information, and motion state of the observation data. This joint method is particularly effective in an environment with large observation noise, can dynamically adjust the observation weights, give priority to more reliable sources, and thus improve the confidence of the estimation results. Finally, it realizes the output of a time and space consistent state sequence of joint dynamic and static obstacles with the observation scene as the unit.

[0032] S2 In the second aspect of the present invention, a group behavior guidance and intelligent vehicle decision-making strategy under a unified scene perspective based on heterogeneous multi-intelligent vehicle fusion perception is provided. It mainly includes group behavior guidance under a unified scene perspective and intelligent vehicle system decision-making based on group behavior guidance, specifically as follows:

[0033] S21 For the group behavior guidance under a unified scene perspective, based on the joint perception processing results of the traffic management area by the cloud, rich traffic motion state data is collected, the motion strategies are aggregated in the cloud, and based on this, the trajectory prediction at the joint scene level of traffic participants is improved, the joint cognition of their behaviors is realized, and a group behavior guidance framework under a unified scene perspective is constructed. Through comprehensive perception and reasonable speculation, the specific steps are as follows:

[0034] Data collection: Using the cloud perception results, widely collect driving data related to various vehicles, including but not limited to vehicle driving trajectories, speed changes, steering behaviors, etc.

[0035] Data preprocessing: Clean and denoise the collected data, and unify it to the same time reference. Adopt a dynamic delay compensation strategy to correct the delay caused by network transmission to ensure the spatio-temporal consistency of all data.

[0036] Prediction Model Construction and Training: A scenario-level trajectory prediction model based on deep learning is adopted. The inputs include historical state variables (such as position and speed), semantic information (such as vehicle type and color), and scene context (such as road network structure and traffic signals). Technologies such as graph neural network (GNN), long short-term memory network (LSTM), or attention mechanism are used to model the dynamic and static interaction relationships between multiple vehicles and map elements in the scene. A supervised learning method is used to minimize the error between the predicted trajectory and the true trajectory, and the probability distribution of the prediction result is output.

[0037] Prediction Model Enhancement: The present invention implements three types of prediction model enhancement strategies, mainly including for different driving scenarios, for different types of traffic participants, and for different driving conditions, as follows:

[0038] For different driving scenarios, the monitoring data collected by the cloud is used to train specific prediction models in units of scenarios, enhancing the adaptability of the model to specific intersections or road sections;

[0039] For different types of traffic participants: The motion characteristics and semantic information of different types of vehicles are integrated into the model, and the trajectory distributions of different types of vehicles (such as buses, private cars, engineering vehicles, etc.) are inferred through joint modeling.

[0040] For different driving conditions: The driving data under different weather conditions (such as sunny, rainy, snowy, etc.) is collected and analyzed, and the model parameters are optimized to ensure that the model can accurately predict the trajectories of non-intelligent vehicles under various environmental conditions.

[0041] The intelligent vehicle collaborative decision-making based on group behavior guidance in S22. On the basis of scenario-level prediction behavior guidance, the present invention realizes the collaborative decision-making of intelligent vehicle groups through low-earth orbit satellite communication. The group collaborative decision-making not only focuses on individual optimality, but also improves the overall traffic efficiency through collaborative cooperation on the basis of fully understanding the group behavior of non-intelligent vehicles, mainly including collaborative decision-making strategy formulation, non-intelligent group influence and decision-making optimization, and real-time adjustment, as follows:

[0042] The collaborative decision-making strategy formulation: The cloud receives the real-time positioning information and target positions of intelligent vehicles, combined with the predicted trajectories of non-intelligent vehicles; The multi-agent reinforcement learning (MARL) algorithm is used to formulate collaborative decision-making strategies. In the MARL framework, each intelligent vehicle acts as an agent, and according to its own goals (such as safe and efficient passage) and environmental perception, it conducts information interaction and cooperation with other intelligent vehicles; By sharing information and negotiating priorities, the traffic flow is jointly optimized. For example, in the intersection scenario, intelligent vehicles determine the passing order through communication negotiation to avoid conflicts. The strategy optimization process can be expressed as:

[0043]

[0044] where π is the policy, γ t is the discount factor, R is the reward function, s t is the state quantity, a t is the action quantity, T is the trajectory time step, is the expected average return that the policy can achieve during long-term operation.

[0045] The impact of non-intelligent vehicles in the game decision-making: During the collaborative decision-making process, intelligent vehicles fully consider the predicted trajectories of non-intelligent vehicles' behaviors and adjust their own decision-making strategies to achieve safe interaction. For example, when it is predicted that a non-intelligent vehicle may suddenly change lanes, the intelligent vehicle decelerates in advance or adjusts its driving route to ensure safe passage. Specifically, the path can be adjusted through dynamic programming methods;

[0046] The decision optimization and real-time adjustment: With the real-time communication and high-precision positioning of low-orbit satellites, the intelligent vehicle group can obtain traffic information and actual driving effects in real time and dynamically optimize the decision-making strategies. In case of emergencies such as traffic accidents and road control, intelligent vehicles can quickly adjust their decision-making strategies to adapt to the new environment. For example, re-planning the driving route or changing the traffic flow direction. At the same time, during the training process, reinforcement learning algorithms (such as Q-learning, policy gradient algorithms) are used to dynamically optimize the parameters of the policy network, making future decisions more adaptable to the actual traffic situation. The optimization goal is to maximize the overall traffic efficiency and safety.

[0047] Based on the above method, the present invention also proposes an intelligent vehicle fusion perception and decision-making system based on low-orbit satellite communication and high-precision positioning, including a perception part and a decision part. The perception part can realize the fusion of heterogeneous perception information in the above S1; the decision part can realize the collaborative decision-making and game in the above S2.

[0048] The beneficial effects of the present invention:

[0049] (1) By combining the high-precision positioning of low-orbit satellites with the perception information fusion strategy of heterogeneous multi-intelligent vehicles, the present invention can effectively improve the accuracy and stability of global perception in complex traffic scenarios. Through real-time and accurate data fusion, it provides reliable scenario-based data support for the collaborative decision-making of intelligent vehicles, reduces decision-making biases caused by perception errors, and thus significantly improves the safety and efficiency of the traffic system.

[0050] (2) By constructing a scenario-level trajectory prediction model based on deep learning and performing personalized optimization for different driving scenarios, traffic participant types, and diverse weather conditions, the model is made to better fit the actual needs. This optimization strategy effectively improves the reliability of vehicle behavior prediction, enhances the vehicle's interaction ability in a dynamic traffic environment, and further improves the decision-making accuracy and response ability of intelligent vehicles in complex traffic scenarios.

[0051] (3) The intelligent vehicle collaborative decision-making mechanism based on the multi-agent reinforcement learning (MARL) algorithm can fully consider the behavior characteristics of various traffic participants and the overall traffic efficiency. Through the collaborative cooperation among multiple agents, the management of traffic flow is optimized, the overall traffic efficiency is improved, traffic congestion and accident risks are reduced, and the traffic system is promoted to develop in a more efficient and intelligent direction.

[0052] (4) Through the real-time collaborative interaction between the cloud and intelligent vehicles, the present invention constructs a game and cooperation system between intelligent vehicles and non-intelligent vehicles. This system provides an innovative technical means for dynamic traffic flow management in complex traffic scenarios, helps to more accurately predict and regulate traffic flow, and ensures a rapid response in case of emergencies or complex traffic situations. It provides important technical support for the comprehensive promotion of intelligent transportation systems and the implementation of autonomous driving technologies, and has broad application prospects and commercial value. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 is a schematic diagram of the key steps of the disclosed embodiment of the present invention;

[0054] Figure 2 is a flowchart of the perception result fusion and process optimization based on low-earth orbit satellite communication and high-precision positioning capabilities of the disclosed embodiment of the present invention;

[0055] Figure 3 is a flowchart of the group behavior guidance and intelligent vehicle collaborative decision-making strategy from the perspective of a unified scenario based on multi-intelligent vehicle fusion perception of the disclosed embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0056] The present invention will be further described below with reference to the accompanying drawings.

[0057] As Figure 1 shown in the schematic diagram of the steps of the embodiment of the present invention, the present invention includes two aspects:

[0058] S1 In the first aspect, the present invention provides a method for post-fusion of perception results and process optimization based on low-earth orbit satellite communication and high-precision positioning capabilities. As Figure 2 shown, the specific content is as follows:

[0059] S11 Information collection: In a complex traffic scenario, the present invention constructs a communication link between the cloud and intelligent vehicles through low-earth orbit satellites. Due to differences in sensor configurations and autonomous driving algorithms among different intelligent vehicles, in order to achieve comprehensive and accurate scene perception, a post-fusion technology is adopted to integrate the perception information of each vehicle. Specifically, each intelligent vehicle collects the perception data of the current scene, including the state parameters of neighboring vehicles (such as speed, acceleration, heading angle, etc.), three-dimensional detection frames (identifying the spatial position and contour of objects), and semantic information (such as vehicle type, color, road signs, etc.). At the same time, the vehicle calculates its own high-precision positioning information (such as position coordinates, heading angle, speed, acceleration, etc.) by receiving multiple groups of low-earth orbit satellite signals and combining with a high-precision positioning algorithm. These state data are compressed and encoded and then quickly uploaded to the cloud through low-earth orbit satellite communication technology to support subsequent data fusion processing.

[0060] Taking an urban crossroads as an example, multiple heterogeneous intelligent vehicles are driving at this traffic intersection, equipped with various sensors such as lidar, cameras, and millimeter-wave radars, continuously detecting the state and semantic information of non-intelligent vehicles around. At the same time, the vehicle uses low-earth orbit satellite signals and ground station signals to calculate its own positioning, speed and other state information. After encoding processing, these perception data and positioning information are transmitted to the cloud in real time through the low-earth orbit satellite link.

[0061] S12 Cloud information processing: After the cloud receives the perception information uploaded by multiple intelligent vehicles, it performs fusion on the target levels based on high-precision positioning, thereby improving the accuracy of complex traffic scene understanding. This process adopts a series of key strategies, including time delay compensation, semantic association matching, and multi-observation joint state trajectory. The specific content is as follows:

[0062] Time delay compensation: Given the dynamic nature of the traffic scene and the delay characteristics of perception information transmission, it is necessary to perform delay compensation on the perception results of each intelligent vehicle. Based on the motion state of the target object, such as position, speed, and acceleration, a motion prediction model is constructed, and combined with the transmission time and the delay time between systems, the motion state of the target object at the current moment is accurately predicted to offset the delay error. Suppose the position of the target vehicle at time t0 is (x0, y0), the speed is the acceleration is After a delay of Δt, at the cloud receiving time t1 = t0 + Δt, taking the uniformly accelerated kinematic formula as an example, its predicted position (x1, y1) can be obtained by the following formula:

[0063]

[0064] The predicted value of speed is:

[0065]

[0066] Through this prediction compensation, the timeliness and accuracy of the perceived information are effectively improved.

[0067] Semantic matching: In the multi-vehicle perception scenario, for the overlapping area of the observation range, it is necessary to accurately match the targets. By utilizing the high-precision positioning ability of low-orbit satellites and combining with traditional multi-sensor fusion matching algorithms (such as the Intersection over Union (IOU) matching strategy), target association can be initially achieved. On this basis, an enhanced matching strategy based on semantic information is incorporated. The extracted semantic information such as color and vehicle type is combined with the vehicle positioning data to construct a joint feature containing spatial position and semantic labels. Through position-based nearest neighbor matching or semantic information-based weighted matching methods, the perception data of each vehicle is matched with the pre-calibrated information in the map. For the same observed target, multi-source information such as color, type, and position is comprehensively considered, and weights are dynamically assigned according to the credibility of each information source for fusion. The calculation method of the joint matching score is as follows:

[0068] Combining the positioning information and semantic information of the vehicle, including color, vehicle type, etc., construct a joint feature vector F = [P, S], where P represents the spatial position feature and S represents the semantic label;

[0069] The matching algorithm adopts a position-based nearest neighbor matching algorithm or a semantic information-based weighted matching method to match the perception data of each vehicle with the information pre-calibrated in the map. The matching degree calculation formula can be expressed as:

[0070] S 匹配 = αIoU + βS 相似度

[0071] where α and β are weight parameters, IoU is the basic matching score based on the intersection over union mentioned above, and S 相似度 is the matching similarity score obtained by encoding and decoding the joint feature vector base F of the neural network:

[0072]

[0073] where f ω is the neural network model, is the semantic joint feature vectors of N vehicles to be observed collected.

[0074] Multi-observation Joint State Estimation: Based on the delay compensation and semantic matching strategy, a multi-observation joint state estimation strategy is introduced to fuse the perception data of multiple intelligent vehicles and other observation sources, forming a unified state estimation of traffic scene targets. Delay compensation provides a unified spatio-temporal benchmark, and semantic matching optimizes the observation consistency. On this basis, the Kalman filter or unscented Kalman filter method is used to incorporate the compensated and matched observation data into the unified state estimation of the target. Taking the intelligent transportation scene as an example, in multi-observation joint state estimation, the Kalman filter plays a core role, which can effectively fuse multi-source observation data and provide accurate system state estimation. Taking the intelligent transportation scene as an example, assume that there are n intelligent vehicles observing a target vehicle, and the system state vector x k contains the state information of the target vehicle such as position, speed, and acceleration, with a dimension of m, that is The observation vector z k,i is the observation value of the target vehicle by the i-th intelligent vehicle, with a dimension of p, that is The state transition matrix F k describes the transition relationship of the system state at adjacent times. For example, in a uniform motion model, if the state vector x k =[x, y, v x , v y T (x and y are position coordinates, v x , v y are velocity components), then:

[0075]

[0076] where Δt is the time interval, indicating that within the time interval Δt, the position changes according to the speed and the speed remains unchanged. The observation matrix H k,i maps the system state to the observation space. For example, if the observation value is only the position information of the target vehicle, then:

[0077]

[0078] The process noise covariance Q k characterizes the uncertainty of the system model itself, such as the uncertainty of state changes caused by factors such as road bumps and minor fluctuations in the power system during vehicle driving; the observation noise covariance R k,i reflects the errors in the observation process, such as the observation deviation caused by sensor accuracy limitations and environmental disturbances (such as the influence of light, rain, and snow on cameras and radars).

[0079] The prediction step of the Kalman filter:

[0080]

[0081] Here ​is the predicted value of the current state based on the previous state estimate. is the covariance matrix of the predicted state, reflecting the uncertainty of the prediction, P k-1 represents the state vector at time k - 1, where the subscripts k and k - 1 denote time k and time k - 1 respectively.

[0082] Update steps:

[0083]

[0084] where, R k,i is the observation noise covariance, K k,i is the Kalman gain, which weighs the weights of the predicted value and the observed value and is dynamically adjusted according to the prediction covariance and the observation covariance. For observation sources with different reliabilities, the covariance is different, and the Kalman gain changes accordingly, making the final state estimate biased towards more reliable information. For example, when the sensor of a certain intelligent vehicle has high precision (small observation noise covariance), its observed value has a greater weight in the fusion.

[0085] Compared with single-source estimation, multi-observation joint state estimation has significant advantages. Single-source estimation only relies on the data of a single vehicle or sensor and is limited by its local perspective and its own errors. For example, a single camera may lose the target or misjudge the position and speed of the target under direct strong light; while multi-observation joint state estimation integrates the information of multiple vehicles and multiple sensors, uses the dynamic weighting mechanism of Kalman filtering, combines the advantages of all parties, suppresses noise, improves the accuracy and stability of state estimation, and thus provides reliable global state information for the decision-making of intelligent vehicles, significantly enhancing the operation efficiency of the intelligent transportation system in complex scenarios.

[0086] S2 Secondly, provide a group behavior guidance and intelligent vehicle collaborative decision-making strategy under the unified scenario perspective based on multi-intelligent vehicle fusion perception, such as Figure 3 shown. The specific content is as follows:

[0087] S21 Group behavior guidance under the unified scenario perspective. After the cloud realizes perception fusion, it further uses these perception results to collect data on various traffic participants and conduct group behavior guidance, aiming to build an intelligent transportation management system under the unified scenario perspective. First, the cloud widely collects driving data involving various vehicles from the real traffic scenario, including information such as speed, acceleration, driving trajectory, and turn signal. These original data are preprocessed, unified to the same time base, and corrected for the delay caused by network transmission through a dynamic delay compensation strategy to ensure that all data has spatio-temporal consistency, providing a high-quality data basis for subsequent model training.

[0088] Based on the preprocessed data, a scene-level trajectory prediction model based on deep learning is constructed. The inputs of this model include multi-source information, covering the historical state of the vehicle (such as speed and position changes over a period of time), semantic information (such as vehicle type, color, usage, etc.), and scene context (such as road network structure, traffic signal status, road slope curvature, etc.). In terms of model architecture selection, in this embodiment, a graph neural network (GNN), a sequence network (such as a long short-term memory network LSTM), and an attention mechanism are adopted to model the interaction relationship between the historical state of the vehicle and multi-modal elements in the traffic scene. For example, the GNN can abstract the traffic scene into a graph structure, where nodes represent vehicles and key road nodes, and edges represent the connection relationships between them. Through the feature transfer and update of nodes and edges, the mutual influence between vehicles can be accurately captured; the LSTM, relying on its advantages in processing time series data, memorizes the historical trajectory features and predicts the future trajectory trend; the attention mechanism can focus on the vehicles or scene areas that have a key impact on trajectory prediction, thereby improving the prediction accuracy. This model outputs the probability distributions of the joint prediction trajectory and the multi-modal prediction trajectory for all non-intelligent vehicles, providing a basis for the group game decision-making and collaborative planning of intelligent vehicles.

[0089] To enable the prediction model to adapt to the complex and changeable traffic environment, leveraging the powerful communication and interaction capabilities of low-earth orbit satellites, it is optimized and enhanced for different driving scenarios, traffic participant types, and driving conditions:

[0090] Regarding different driving scenarios, the cloud collects the driving data of non-intelligent vehicles from different monitored intersections (such as urban arterial road intersections, suburban curves, highway ramps, etc.), and conducts targeted model training for the characteristics of each scenario to ensure that the model output can fit the unique traffic flow characteristics and driving trajectory patterns of each intersection;

[0091] For traffic participant types, the model integrates motion characteristics and semantic information. For example, considering the characteristics of buses that frequently stop at stations and start slowly, and the characteristics of engineering vehicles with relatively fixed driving routes and speed limitations, through joint modeling of position and vehicle type, the trajectory distribution of specific types of vehicles can be accurately inferred;

[0092] In the dimension of different driving conditions, the cloud collects driving behavior data under various weather conditions such as sunny, rainy, snowy, and foggy days, and validates and tests the model to ensure that it can accurately predict the trajectories of non-intelligent vehicles under complex meteorological conditions.

[0093] S22 Collaborative Decision-making for Intelligent Vehicles Based on Group Behavior Guidance. Based on the vehicle scenario-level predictive behavior guidance in specific driving scenarios, the present invention aims to achieve collaborative decision-making for all intelligent vehicle groups connected through low-earth orbit satellite communication links. The decision-making of intelligent vehicles should not be limited to individual optimization, but through collaborative cooperation, on the basis of fully understanding the behavior of other vehicle groups, to improve the overall traffic efficiency. The specific strategies are as follows:

[0094] Formulation of Collaborative Decision-making Strategies: Based on scenario-level trajectory prediction, the present invention realizes the collaborative decision-making of all intelligent vehicles in the currently monitored scenario through the cloud, so as to break through the limitation of individual optimization and maximize the overall traffic efficiency. The specific process is as follows:

[0095] The cloud receives the positioning information and end target positions uploaded in real time by all intelligent vehicles, and uses the multi-agent reinforcement learning (MARL) algorithm to formulate the collaborative decision-making strategies for intelligent vehicles. In the MARL framework, each intelligent vehicle is regarded as an agent, and based on its own goals (such as safety, efficient travel, and optimal energy consumption) and perception of the environment, it interacts and cooperates with other agents. For example, in the intersection scenario, intelligent vehicles negotiate through low-earth orbit satellite communication to determine their respective passing orders, comprehensively consider factors such as vehicle speed, intersection distance, and turning direction, reasonably allocate road rights, avoid collision conflicts, and ensure smooth traffic at the intersection.

[0096] During the collaborative decision-making process, intelligent vehicles fully consider the behavior and predicted trajectories of non-intelligent vehicles and regard them as key decision variables. By deeply analyzing the possible behaviors of non-intelligent vehicles (such as sudden lane changes, hard brakes, illegal turns, etc.) and their potential impacts on traffic flow, intelligent vehicles timely adjust their own decision-making strategies to achieve safe and harmonious interaction with non-intelligent vehicles. For example, when it is predicted that a non-intelligent vehicle may suddenly cut into the lane, the intelligent vehicle will decelerate in advance or slightly adjust its driving route to reserve a safe buffer space to ensure driving safety.

[0097] With the real-time communication and high-precision positioning capabilities of low-earth orbit satellites, intelligent vehicle groups can continuously optimize their decision-making strategies based on real-time traffic information (such as sudden situations like traffic accidents, temporary road controls, and traffic flow mutations) and actual driving effects. When encountering sudden situations, intelligent vehicles quickly adjust their decisions and rapidly adapt to the new traffic environment. At the same time, through continuous interaction with other intelligent vehicles, the collaborative decision-making mechanism is continuously improved, gradually enhancing the overall traffic efficiency and safety. During the training process, classic algorithms in reinforcement learning algorithms (such as Q-learning, policy gradient algorithms, etc.) are used to update and optimize the decision-making strategies according to the dynamically changing driving situations, adjust the parameters in the policy network, so that future decisions can better fit the actual traffic conditions, and realize the self-learning and self-adaptive evolution of the intelligent transportation system. The specific process of the MARL algorithm is as follows:

[0098] First, in the initialization phase, each intelligent vehicle, as an independent agent, sets its own initial state including vehicle position, speed, orientation, surrounding environment perception information, etc., and at the same time sets the initial strategy This strategy determines the next action based on the current state of the vehicle, such as accelerating, decelerating, turning, etc. The cloud, as a coordination center, constructs a global environmental state containing the state information of all intelligent vehicles

[0099] At each decision-making moment t, each intelligent vehicle, according to its current own state and strategy selects an action and uploads the action information to the cloud in real time through low-earth orbit satellite communication. For example, intelligent vehicle A is at an intersection, and there is a non-intelligent vehicle decelerating ahead. According to its own strategy, it decides to decelerate and slightly adjust its direction to the right, and uploads this action information.

[0100] After the cloud receives the actions of all intelligent vehicles, it updates the global environmental state S t+1 . According to traffic rules, road conditions, and the actions of other vehicles, it calculates the immediate reward for each intelligent vehicle The reward function comprehensively considers various factors, such as safety factors (high reward for avoiding collisions, penalty for approaching dangerous distances), efficiency factors (reward for passing quickly, penalty for frequent starts and stops), and cooperation factors (reward for good cooperation with other intelligent vehicles, penalty for conflicts and interference).

[0101] Subsequently, each intelligent vehicle, based on the new global environmental state S t+1 and the obtained immediate reward uses the policy gradient algorithm to update its own policy During the training process, the above decision-making, interaction, and update steps are repeated multiple times. The intelligent vehicles continuously optimize their own strategies and gradually learn to cooperate with other intelligent vehicles in complex traffic scenarios to improve the overall traffic efficiency.

[0102] The series of detailed descriptions listed above are only specific descriptions of the feasible implementation modes of the present invention, and they are not intended to limit the protection scope of the present invention. Any equivalent methods or changes that do not deviate from the technology created by the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent vehicle fusion perception and decision-making method based on low-orbit satellite communication and high-precision positioning, characterized in that, Including: Fusion of S1 heterogeneous perception information: By integrating the perception data of multi-source heterogeneous intelligent vehicles, a comprehensive understanding of the traffic scenario is achieved; S2 Cooperative decision-making and game: Based on the comprehensive understanding of the traffic scenario, the game decision-making between intelligent vehicles and non-intelligent vehicles is optimized, and at the same time, cooperative decision-making within the intelligent vehicle system is realized.

2. The intelligent vehicle fusion perception and decision-making method based on low-orbit satellite communication and high-precision positioning according to claim 1, characterized in that The specific implementation of the above S1 includes: S11 Acquisition and transmission of environmental perception data of heterogeneous intelligent vehicles; Specifically: In a complex traffic scenario, a communication link between the cloud and each intelligent vehicle is constructed using low-earth orbit satellites. For the heterogeneous differences in sensor configurations and algorithms of different intelligent vehicles, a post-fusion method is adopted to integrate perception information; The information acquisition content includes perception data acquisition, data encoding, and transmission: The perception data acquisition includes the perception data of heterogeneous intelligent vehicles: adjacent vehicle status, three-dimensional detection boxes, semantic information; and the high-precision vehicle status information calculated through the signals of multiple low-earth orbit satellites and ground monitoring stations: position, heading angle, speed, acceleration; The data encoding and transmission means that the collected data is compressed and encoded and then uploaded to the cloud server for storage and processing through the communication capabilities of low-earth orbit satellites; Efficient encoding means are adopted during the transmission process to reduce bandwidth while ensuring the real-time and integrity of the transmission.

3. The intelligent vehicle fusion perception and decision-making method based on low-orbit satellite communication and high-precision positioning according to claim 2, characterized in that, The specific implementation of the above S1 also includes: S12 Information processing by the cloud server; Specifically: In the cloud server, the perception status information collected and published by heterogeneous multi-intelligent vehicles in the supervision scenario is received through low-earth orbit satellite communication, and target-level fusion is achieved based on high-precision positioning to realize accurate scenario understanding in a highly uncertain traffic scenario; The processing strategies of the cloud include time delay compensation, semantic association matching, and multi-observation joint state trajectory estimation: The time delay compensation is based on the multi-source perception results of heterogeneous intelligent vehicles to perform delay compensation for all observed targets to be estimated; The semantic association matching means that in the observed scenario, for multiple intelligent vehicles with perception capabilities, target matching is performed for the overlapping parts of their observation ranges; The multi-observation joint state estimation, based on time delay compensation and semantic association matching, introduces a multi-observation joint state estimation strategy to fuse the perception data of multiple intelligent vehicles and other observation sources to form a unified state estimation of the targets in the traffic scenario; Based on time delay compensation, the unified spatio-temporal reference is established, and the target association algorithm based on semantic information matching optimizes the observation consistency of the state estimation; On this basis, through Kalman filtering or unscented Kalman filtering, all compensated and matched observation data are fused into the unified state estimation of the target.

4. The intelligent vehicle fusion perception and decision-making method based on low-orbit satellite communication and high-precision positioning according to claim 3, wherein The specific design of the time delay compensation is as follows: Construct a motion prediction model: Based on the motion state of the target object, a motion prediction model for each vehicle is constructed, using a uniform motion model or a uniformly accelerated motion model; Calculate the delay time: Based on the data transmission delay and the perception link delay, calculate the possible motion state delay time Δt of the target object; State prediction: Use the motion prediction model to predict the current accurate position of the target object to obtain the compensated state quantity.

5. The intelligent vehicle fusion perception and decision-making method based on low-earth orbit satellite communication and high-precision positioning according to claim 4, characterized in that, The semantic association matching includes two parts: basic matching based on the intersection over union (IoU) and semantic feature enhanced matching, which are specifically as follows: For the basic matching based on the intersection over union (IoU), the high-precision positioning ability of the low-orbit satellite is used. According to the spatial positioning of the sensing carrier, three-dimensional spatial projection is performed on the perceived target. Taking the driving scenario as a unit, the intersection over union (IoU) matching strategy is adopted to achieve basic spatial matching, and a preliminary matching result of heterogeneous intelligent vehicle perception is constructed; The semantic feature enhanced matching, that is, using high-level semantic information to achieve high-order joint matching, includes target association feature extraction, matching algorithm design, and dynamic weight adjustment, which are specifically as follows: For the association feature extraction, combining the positioning information and semantic information of the vehicle, including color and vehicle type, a joint feature vector F = [P, S] is constructed, where P represents the spatial position feature and S represents the semantic label; For the matching algorithm design, a nearest neighbor matching algorithm based on position or a weighted matching method based on semantic information is adopted to match the sensing data of each vehicle with the information pre-calibrated in the map. The matching degree calculation formula can be expressed as: S 匹配 = α IoU + β S 相似度 where α and β are weight parameters, IoU is the above-mentioned basic matching score based on the intersection over union, and S 相似度 is the matching similarity score obtained by encoding and decoding the joint feature vector basis F based on a neural network: where f ω is a neural network model, is the semantic joint feature vector collected from N vehicles to be observed; For the dynamic weight adjustment, the weights of each information source are adjusted according to the environmental conditions. When the light is good, the color information has high credibility and a larger weight is given; in bad weather, the position information is more reliable and the weights are adjusted accordingly.

6. The intelligent vehicle fusion perception and decision-making method based on low-earth orbit satellite communication and high-precision positioning according to claim 1, characterized in that The implementation of S2 includes: S21 Group behavior guidance under a unified scene perspective: Based on the joint perception processing results of the traffic management area by the cloud, rich traffic movement state data is collected, the movement strategies of the cloud are aggregated, and based on this, the trajectory prediction at the joint scene level of traffic participants is improved, the joint cognition of their behaviors is realized, a group behavior guidance framework under a unified scene perspective is constructed, and comprehensive perception and reasonable speculation are achieved. The specific steps are as follows: Data collection: Using the cloud perception results, widely collect driving data related to various vehicles, including vehicle driving trajectories, speed changes, and steering behaviors; Data preprocessing: Clean and denoise the collected data, and unify it to the same time base. Adopt a dynamic delay compensation strategy to correct the delay caused by network transmission to ensure the spatio-temporal consistency of all data; Construction and training of the prediction model: Adopt a scene-level trajectory prediction model based on deep learning. The inputs include historical state variables, semantic information, and scene context. Use a graph neural network (GNN), long short-term memory network (LSTM), or attention mechanism to model the dynamic and static interaction relationships between multiple vehicles and map elements in the scene; Adopt a supervised learning method to minimize the error between the predicted trajectory and the real trajectory, and output the probability distribution of the prediction result; Enhancement of the prediction model: The present invention realizes three types of prediction model enhancement strategies, including for different driving scenarios, for different types of traffic participants, and for different driving conditions, which are specifically as follows: For different driving scenarios, using the monitoring data collected by the cloud, train specific prediction models in units of scenarios to enhance the adaptability of the model to specific intersections or sections; Regarding different types of traffic participants: Incorporate the motion characteristics and semantic information of different types of vehicles into the model, and infer the trajectory distribution of different types of vehicles through joint modeling. Regarding different driving conditions: Collect and analyze driving data under different weather conditions, and optimize the model parameters to ensure that the model can accurately predict the trajectories of non-intelligent vehicles under various environmental conditions.

7. The intelligent vehicle fusion perception and decision-making method based on low-earth orbit satellite communication and high-precision positioning according to claim 6, characterized in that The implementation of S2 also includes: S22 Intelligent vehicle collaborative decision-making based on group behavior guidance; On the basis of scenario-level prediction behavior guidance, through low-earth orbit satellite communication, realize the collaborative decision-making of intelligent vehicle groups; The group collaborative decision-making not only focuses on individual optimization, but also improves the overall traffic efficiency through collaborative cooperation on the basis of fully understanding the group behavior of non-intelligent vehicles, including the formulation of collaborative decision-making strategies, the judgment of the impact of non-intelligent groups, decision optimization and real-time adjustment, as follows: The formulation of the collaborative decision-making strategy: The cloud receives the real-time positioning information and target positions of intelligent vehicles, and combines the predicted trajectories of non-intelligent vehicles; Use the Multi-Agent Reinforcement Learning (MARL) algorithm to formulate the collaborative decision-making strategy. The judgment of the impact of non-intelligent groups: During the collaborative decision-making process, intelligent vehicles fully consider the predicted trajectories of non-intelligent vehicles' behaviors and adjust their own decision-making strategies to achieve safe interaction. Decision optimization and real-time adjustment: With the help of the real-time communication and high-precision positioning of low-earth orbit satellites, intelligent vehicle groups can obtain traffic information and actual driving effects in real time, and dynamically optimize decision-making strategies; In case of emergencies such as traffic accidents and road control, intelligent vehicles can quickly adjust decision-making strategies to adapt to the new environment, including re-planning driving routes or changing traffic flows.

8. The intelligent vehicle fusion perception and decision-making method based on low-earth orbit satellite communication and high-precision positioning according to claim 7, wherein The specific steps of using the MARL algorithm to formulate the collaborative decision-making strategy are as follows: Each intelligent vehicle acts as an agent, and according to its own goals and environmental perception, conducts information interaction and cooperation with other intelligent vehicles; By sharing information and negotiating priorities, jointly optimize the traffic flow; For example, in the scenario of an intersection, intelligent vehicles determine the passing order through communication negotiation to avoid conflicts. The process of optimizing this strategy can be expressed as: where π is the policy, γ is the discount factor, R is the reward function, s t is the state quantity, and a t is the action quantity.

9. The intelligent vehicle fusion perception and decision-making method based on low-earth orbit satellite communication and high-precision positioning according to claim 7, characterized in that During the training process of the decision-making network, use the reinforcement learning algorithm to dynamically optimize the parameters of the decision-making network to make future decisions more adaptable to the actual traffic situation. The optimization goal is to maximize the overall traffic efficiency and safety.

10. An intelligent vehicle fusion perception and decision-making system based on low-earth orbit satellite communication and high-precision positioning, characterized in that, It includes a perception part and a decision part. The perception part can realize the fusion of heterogeneous perception information described in claims 2-5; The decision part can realize the collaborative decision-making and game described in claims 6-9.

Images