Intelligent obstacle avoidance method and system based on intelligent networked automobile

Abstract

The invention discloses an intelligent obstacle avoidance method and system based on an intelligent networked automobile, and relates to the field of intelligent networked automobiles, and the method comprises the steps: constructing a three-dimensional environment map through a point cloud fusion technology, and determining the position information of the intelligent networked automobile in the three-dimensional environment map through a multi-sensor fusion positioning algorithm; a deep learning algorithm is used to identify spatial distribution of static obstacles and motion features of dynamic obstacles, and a neural network is used to predict motion trails of the dynamic obstacles; generating an obstacle avoidance path by using a path planning algorithm, and optimizing the obstacle avoidance path in combination with the multi-dimensional scene parameters to obtain an obstacle avoidance strategy; a layered simulation framework is constructed, a coupling model is established in a digital twin platform, obstacle avoidance testing is performed by using model prediction control, and an obstacle avoidance strategy is optimized based on a simulation test result. According to the invention, the sensing capability and the response accuracy of the intelligent connected automobile in a complex traffic environment are greatly enhanced.

Description

Technical Field

[0001] This invention relates to the field of intelligent connected vehicles, and more specifically, to an intelligent obstacle avoidance method and system based on intelligent connected vehicles. Background Technology

[0002] With the acceleration of urbanization and the rapid increase in the number of vehicles, the incidence of traffic accidents has also risen accordingly. Among these accidents, a considerable proportion are caused by drivers' failure to detect obstacles in time or to take appropriate measures to avoid them. Traditional vehicles rely on the driver's visual, auditory, and other sensory abilities, as well as reaction speed, to identify and avoid obstacles on the road. However, this reliance on human drivers has certain limitations, such as limited visibility at night and slowed reaction time due to driver fatigue.

[0003] In recent years, the development of intelligent connected vehicle technology has provided new ideas and technical means to solve the above problems. Intelligent connected vehicles can not only perceive the surrounding environment in real time through onboard sensors (such as radar, lidar, cameras, etc.), but also exchange information with other vehicles, infrastructure and other traffic participants through vehicle networking technology, so as to achieve more accurate environmental perception and decision support.

[0004] Despite significant progress in intelligent obstacle avoidance technology, complex and ever-changing real-world road conditions and different driving scenarios (such as urban roads, highways, and adverse weather conditions) place higher demands on obstacle avoidance algorithms, and these demands vary from scenario to scenario. However, most existing obstacle avoidance algorithms tend to simplify the process or are only optimized for specific situations, making them inadequate when dealing with complex and ever-changing traffic environments.

[0005] There are currently no effective solutions to the problems in the relevant technologies. Summary of the Invention

[0006] In view of the problems in related technologies, this invention proposes an intelligent obstacle avoidance method and system based on intelligent connected vehicles to overcome the above-mentioned technical problems existing in the existing related technologies.

[0007] Therefore, the specific technical solution adopted by the present invention is as follows:

[0008] According to one aspect of the present invention, an intelligent obstacle avoidance method based on intelligent connected vehicles is provided, comprising:

[0009] S1. Collect multimodal environmental data, construct a three-dimensional environmental map using point cloud fusion technology, and determine the location information of the intelligent connected vehicle in the three-dimensional environmental map through a multi-sensor fusion positioning algorithm;

[0010] S2. Based on the constructed 3D environment map, deep learning algorithms are used to identify the spatial distribution of static obstacles and the motion characteristics of dynamic obstacles, and the motion trajectory of dynamic obstacles is predicted through neural networks.

[0011] S3. Based on the location information of the intelligent connected vehicle, the spatial distribution of static obstacles and the predicted trajectory of dynamic obstacles, an obstacle avoidance path is generated using a path planning algorithm. The obstacle avoidance path is then optimized by combining multi-dimensional scene parameters to obtain an obstacle avoidance strategy.

[0012] S4. Construct a hierarchical simulation framework and establish a coupled model in the digital twin platform. Use model predictive control to conduct obstacle avoidance tests and optimize the obstacle avoidance strategy based on the simulation test results.

[0013] Optionally, multimodal environmental data is collected, a 3D environmental map is constructed using point cloud fusion technology, and the location information of the intelligent connected vehicle in the 3D environmental map is determined through a multi-sensor fusion positioning algorithm, including:

[0014] S11. Collect multimodal environmental data through several types of sensors, and perform noise reduction, correction and spatiotemporal alignment processing on the multimodal environmental data;

[0015] S12. Extract point cloud features from the processed multimodal environment data, and use point cloud fusion technology to perform multi-frame registration and global optimization in order to construct a three-dimensional environment map.

[0016] S13. Based on the initial pose information, Kalman filtering is used to fuse multimodal environmental data, and a multi-sensor fusion positioning algorithm is used to perform real-time map matching to determine the real-time location information of the intelligent connected vehicle in the three-dimensional environment map.

[0017] Optionally, based on the constructed 3D environment map, deep learning algorithms are used to identify the spatial distribution of static obstacles and the motion characteristics of dynamic obstacles, and neural networks are used to predict the motion trajectories of dynamic obstacles, including:

[0018] S21. Based on the constructed 3D environment map, the point cloud data is preliminarily processed using the point cloud semantic segmentation algorithm in deep learning to distinguish different categories of static obstacles.

[0019] S22. Use the Euclidean clustering algorithm to perform secondary processing on the preliminary processing results to generate static obstacle point cloud clusters containing semantic labels and obtain the spatial distribution of static obstacles.

[0020] S23. Apply target detection algorithms to point cloud data to detect and classify dynamic obstacles, and combine spatiotemporal convolutional networks to extract the motion features of dynamic obstacles.

[0021] S24. Construct a multi-task learning framework, combine the motion characteristics and time series data of dynamic obstacles, and use neural networks to predict the motion trajectory of dynamic obstacles.

[0022] Optionally, based on the location information of the intelligent connected vehicle, the spatial distribution of static obstacles, and the predicted trajectory of dynamic obstacles, an obstacle avoidance path is generated using a path planning algorithm. This path is then optimized using multi-dimensional scene parameters to obtain an obstacle avoidance strategy, including:

[0023] S31. Based on the spatial distribution of static obstacles, obtain the expansion region of static obstacles and set the corresponding safe distance threshold.

[0024] S32. Based on the expansion region and the safety distance threshold, the three-dimensional environment map is divided into different regions to construct a static obstacle constraint map;

[0025] S33. Using the location information of the intelligent connected vehicle as the starting point and the target location as the ending point, combined with the static obstacle constraint diagram, the A* algorithm is used to perform path planning and generate a global path.

[0026] S34. Using the predicted trajectory of dynamic obstacles as a local time-varying constraint, the global path is locally reconstructed using the fast expanding random tree algorithm to generate an obstacle avoidance path.

[0027] S35. Obtain multi-dimensional scene parameters through vehicle-to-everything (V2X) technology, and use fuzzy logic reasoning algorithm to map the multi-dimensional scene parameters into path optimization indicators. Adjust the spatiotemporal parameters of the obstacle avoidance path based on the path optimization indicators to generate an obstacle avoidance strategy.

[0028] Optionally, based on the spatial distribution of static obstacles, the expansion region of the static obstacles is obtained, and a corresponding safe distance threshold is set, including:

[0029] S311. Based on the spatial distribution of static obstacles, the point cloud data of static obstacles is reconstructed into a triangular mesh model, and the triangular mesh model is discretized.

[0030] S312. Based on the discretization results, calculate the shortest distance from each triangular mesh vertex to the static obstacle, and establish a symbolic distance field;

[0031] S313. Based on the established symbolic distance field, segmentation is performed using a preset distance threshold to define the area within a preset range of the static obstacle as the expansion region.

[0032] S314. Set the corresponding safety distance threshold according to the expansion area and the specifications and dimensions of the intelligent connected vehicle.

[0033] Optionally, different areas include: free areas, restricted areas, and restricted areas.

[0034] Optionally, the predicted trajectory of dynamic obstacles is used as a local time-varying constraint, and the global path is locally reconstructed using a fast expanding random tree algorithm to generate obstacle avoidance paths, including:

[0035] S341. Based on the predicted trajectory of the dynamic obstacle, the four-dimensional space is rasterized and the dynamic obstacle is assigned a spatiotemporal occupancy probability to construct a spatiotemporal occupancy map.

[0036] S342. Combining the spatiotemporal occupancy graph with the fast expanding random tree algorithm, the global path is expanded, and the spatiotemporal occupancy probability of the newly expanded node to the global path is queried through collision detection.

[0037] S343. Based on the collision detection results, calculate the overlap between the global path and the predicted trajectory of the dynamic obstacle, and determine whether local path reconstruction is needed by combining the preset overlap threshold.

[0038] S344. If the overlap is not less than the preset overlap threshold, local path reconstruction is triggered to generate a new local path, which is then spliced ​​with the global path to generate an obstacle avoidance path; otherwise, the global path is retained and an obstacle avoidance path is generated.

[0039] Optionally, a hierarchical simulation framework is constructed, and a coupled model is established in the digital twin platform. Obstacle avoidance testing is conducted using model predictive control, and the obstacle avoidance strategy is optimized based on the simulation test results, including:

[0040] S41. Based on the dynamic model of intelligent connected vehicles, and combining the static obstacle constraint diagram and the dynamic obstacle spatiotemporal occupancy diagram, construct a hierarchical simulation framework.

[0041] S42. In the digital twin simulation platform, based on the hierarchical simulation framework, construct a coupled model of intelligent connected vehicles and the environment, and define the control input of the coupled model;

[0042] S43. Based on the coupled model, the model predictive control algorithm is used to track the obstacle avoidance trajectory of intelligent connected vehicles, and simulation tests are carried out in several different scenarios.

[0043] S44. By analyzing the simulation test results, the obstacle avoidance strategy is evaluated. Based on the evaluation results, the obstacle avoidance strategy is optimized using a reinforcement learning algorithm.

[0044] Optionally, based on the coupled model, model predictive control algorithms are used to track the obstacle avoidance trajectory of intelligent connected vehicles, and simulation tests are conducted in several different scenarios, including:

[0045] S431. Based on the coupling model, set the initial state and target point sequence of the intelligent connected vehicle, and load different simulation scenarios;

[0046] S432. Based on the generated obstacle avoidance strategy, set the optimization objective and corresponding constraints, and use the model predictive control algorithm to calculate the optimal control input;

[0047] S433. Based on the optimal control input, execute obstacle avoidance strategies in simulation scenarios with different loads, perform obstacle avoidance trajectory tracking, and record performance data in each simulation process.

[0048] According to another aspect of the present invention, an intelligent obstacle avoidance system based on intelligent connected vehicles is also provided, comprising:

[0049] The vehicle location determination module is used to collect multimodal environmental data, construct a three-dimensional environmental map using point cloud fusion technology, and determine the location information of the intelligent connected vehicle in the three-dimensional environmental map through a multi-sensor fusion positioning algorithm.

[0050] The obstacle recognition module is used to identify the spatial distribution of static obstacles and the motion characteristics of dynamic obstacles based on the constructed 3D environment map using deep learning algorithms, and to predict the motion trajectory of dynamic obstacles through neural networks.

[0051] The obstacle avoidance strategy generation module is used to generate an obstacle avoidance path based on the location information of the intelligent connected vehicle, the spatial distribution of static obstacles and the predicted trajectory of dynamic obstacles, using a path planning algorithm. The obstacle avoidance path is then optimized by combining multi-dimensional scene parameters to obtain an obstacle avoidance strategy.

[0052] The obstacle avoidance strategy optimization module is used to build a hierarchical simulation framework and establish a coupled model in the digital twin platform. It uses model predictive control to conduct obstacle avoidance tests and optimizes the obstacle avoidance strategy based on the simulation test results.

[0053] The beneficial effects of the present invention are:

[0054] 1. This invention integrates multimodal sensor fusion and point cloud mapping technologies to achieve three-dimensional modeling of the surrounding environment. At the same time, combined with multi-sensor positioning algorithms, it can accurately obtain the vehicle's location information, providing a solid foundation for subsequent path planning and obstacle avoidance strategies, and greatly enhancing the perception capabilities and reaction accuracy of intelligent connected vehicles in complex traffic environments.

[0055] 2. By combining deep learning-based obstacle recognition and trajectory prediction technologies, this invention enables intelligent connected vehicles to pre-perceive and assess potential risks in dynamic environments. This allows vehicles to optimize path planning and obstacle avoidance strategies based on multi-dimensional environmental parameters, ensuring not only the safety performance of vehicles in complex and dynamic environments but also improving traffic efficiency and guaranteeing that intelligent connected vehicles can drive smoothly and efficiently under various road conditions.

[0056] 3. By utilizing a hierarchical simulation framework and a digital twin platform, this invention enables large-scale, multi-scenario obstacle avoidance strategy testing in a virtual environment. This not only enhances the adaptability of the obstacle avoidance system but also improves the robustness of control, allowing intelligent connected vehicles to operate stably and effectively avoid obstacles under various conditions. Attached Figure Description

[0057] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0058] Figure 1 This is a flowchart of an intelligent obstacle avoidance method based on an intelligent connected vehicle according to an embodiment of the present invention;

[0059] Figure 2 This is a schematic diagram of a smart obstacle avoidance system based on an intelligent connected vehicle according to an embodiment of the present invention.

[0060] In the picture:

[0061] 1. Vehicle position determination module; 2. Obstacle recognition module; 3. Obstacle avoidance strategy generation module; 4. Obstacle avoidance strategy optimization module. Detailed Implementation

[0062] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention.

[0063] According to an embodiment of the present invention, an intelligent obstacle avoidance method and system based on intelligent connected vehicles is provided.

[0064] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, the intelligent obstacle avoidance method based on intelligent connected vehicles according to an embodiment of the present invention includes:

[0065] S1. Collect multimodal environmental data, construct a three-dimensional environmental map using point cloud fusion technology, and determine the location information of the intelligent connected vehicle in the three-dimensional environmental map through a multi-sensor fusion positioning algorithm.

[0066] Preferably, the process of collecting multimodal environmental data, constructing a three-dimensional environmental map using point cloud fusion technology, and determining the location information of the intelligent connected vehicle in the three-dimensional environmental map using a multi-sensor fusion positioning algorithm includes:

[0067] S11. Collect multimodal environmental data through several types of sensors, and perform noise reduction, correction and spatiotemporal alignment processing on the multimodal environmental data;

[0068] S12. Extract point cloud features from the processed multimodal environment data, and use point cloud fusion technology to perform multi-frame registration and global optimization in order to construct a three-dimensional environment map.

[0069] S13. Based on the initial pose information, Kalman filtering is used to fuse multimodal environmental data, and a multi-sensor fusion positioning algorithm is used to perform real-time map matching to determine the real-time location information of the intelligent connected vehicle in the three-dimensional environment map.

[0070] S2. Based on the constructed 3D environment map, deep learning algorithms are used to identify the spatial distribution of static obstacles and the motion characteristics of dynamic obstacles, and neural networks are used to predict the motion trajectory of dynamic obstacles.

[0071] Preferably, based on the constructed 3D environment map, deep learning algorithms are used to identify the spatial distribution of static obstacles and the motion characteristics of dynamic obstacles, and neural networks are used to predict the motion trajectory of dynamic obstacles, including:

[0072] S21. Based on the constructed 3D environment map, the point cloud data is preliminarily processed using the point cloud semantic segmentation algorithm in deep learning to distinguish different categories of static obstacles.

[0073] S22. Use the Euclidean clustering algorithm to perform secondary processing on the preliminary processing results to generate static obstacle point cloud clusters containing semantic labels and obtain the spatial distribution of static obstacles.

[0074] S23. Apply target detection algorithms to point cloud data to detect and classify dynamic obstacles, and combine spatiotemporal convolutional networks to extract the motion features of dynamic obstacles.

[0075] S24. Construct a multi-task learning framework, combine the motion characteristics and time series data of dynamic obstacles, and use neural networks to predict the motion trajectory of dynamic obstacles.

[0076] It should be further explained that, based on the constructed 3D environment map, a specific implementation example of using deep learning algorithms to identify the spatial distribution of static obstacles and the motion characteristics of dynamic obstacles, and using neural networks to predict the motion trajectory of dynamic obstacles, is as follows:

[0077] 1. Use LiDAR to collect point clouds of roads in the park, label 10 types of static obstacles (such as curbs, traffic cones, and parking bollards), and use PointNet++ for semantic segmentation. The input point cloud density is 0.1m. 3The algorithm outputs the category label for each point; it applies KD-Tree to accelerate neighborhood search for similar point clouds, setting the search radius r = 0.5m; it recursively merges points that meet the distance threshold to generate point cloud clusters, calculates the centroid and bounding box size of each cluster, and achieves an average clustering time of 0.23 seconds / frame (100,000 point clouds) and a clustering accuracy of 91.5% (compared to manual annotation). For example: Cluster 1 (traffic cone), centroid (3.2, 1.5, 0), bounding box size (0.8m, 0.8m, 2.0m); Cluster 2 (curb), centroid (-1.0, -2.3, 0), bounding box size (5.0m, 0.3m, 0.2m).

[0078] 2. Point clouds are converted into pseudo images (resolution 0.16 m2 / pixel) using PointPillars, and features are extracted using the ResNet-18 backbone network to detect dynamic targets such as pedestrians and vehicles. Combine three consecutive frames of point clouds and calculate velocity and acceleration through spatiotemporal convolution. The velocity estimation error does not exceed 0.3 m / s (compared to the true GPS value).

[0079] 3. LSTM encoding of the dynamic target's historical trajectory (5 seconds, 10Hz) is used. A branch network predicts the target's position in the next 3 seconds (regression task) and classifies its behavior (e.g., going straight, turning). Smooth L1 Loss is used for position prediction, and cross-entropy loss is used for behavior classification. The average displacement error is 0.52m (3-second prediction), and the final displacement error is 1.03m (3-second prediction). For example, a logistics channel in a park (100m long, 8m wide) contains the following obstacles: static (curb, traffic cones), dynamic (pedestrian speed 1m / s, forklift speed 2m / s). Static obstacle avoidance involves identifying the curb 15 meters in advance and adjusting the path deviation to ≤0.2m. Dynamic obstacle avoidance involves predicting the position of a suddenly cutting forklift 3 seconds later, planning a detour trajectory, and maintaining a minimum safe distance of 1.5m.

[0080] S3. Based on the location information of the intelligent connected vehicle, the spatial distribution of static obstacles, and the predicted trajectory of dynamic obstacles, an obstacle avoidance path is generated using a path planning algorithm. The obstacle avoidance path is then optimized by combining multi-dimensional scene parameters to obtain an obstacle avoidance strategy.

[0081] Preferably, based on the location information of the intelligent connected vehicle, the spatial distribution of static obstacles, and the predicted trajectory of dynamic obstacles, an obstacle avoidance path is generated using a path planning algorithm. This path is then optimized using multi-dimensional scene parameters to obtain an obstacle avoidance strategy, including:

[0082] S31. Based on the spatial distribution of static obstacles, obtain the expansion area of ​​static obstacles and set the corresponding safe distance threshold.

[0083] Preferably, based on the spatial distribution of static obstacles, the expansion region of the static obstacles is obtained, and a corresponding safe distance threshold is set, including:

[0084] S311. Based on the spatial distribution of static obstacles, the point cloud data of static obstacles is reconstructed into a triangular mesh model, and the triangular mesh model is discretized.

[0085] S312. Based on the discretization results, calculate the shortest distance from each triangular mesh vertex to the static obstacle, and establish a symbolic distance field;

[0086] S313. Based on the established symbolic distance field, segmentation is performed using a preset distance threshold to define the area within a preset range of the static obstacle as the expansion region.

[0087] S314. Set the corresponding safety distance threshold according to the expansion area and the specifications and dimensions of the intelligent connected vehicle.

[0088] S32. Based on the expansion region and the safety distance threshold, the three-dimensional environment map is divided into different regions to construct a static obstacle constraint map.

[0089] Preferably, the different areas include: free areas, restricted areas, and restricted areas.

[0090] S33. Using the location information of the intelligent connected vehicle as the starting point and the target location as the ending point, and combining the static obstacle constraint diagram, the A* algorithm is used to perform path planning and generate a global path.

[0091] S34. Using the predicted trajectory of dynamic obstacles as a local time-varying constraint, the global path is locally reconstructed using the fast extended random tree algorithm to generate an obstacle avoidance path.

[0092] Preferably, the predicted trajectory of dynamic obstacles is used as a local time-varying constraint, and the global path is locally reconstructed using a fast expanding random tree algorithm to generate an obstacle avoidance path, including:

[0093] S341. Based on the predicted trajectory of the dynamic obstacle, the four-dimensional space is rasterized and the dynamic obstacle is assigned a spatiotemporal occupancy probability to construct a spatiotemporal occupancy map.

[0094] S342. Combining the spatiotemporal occupancy graph with the fast expanding random tree algorithm, the global path is expanded, and the spatiotemporal occupancy probability of the newly expanded node to the global path is queried through collision detection.

[0095] S343. Based on the collision detection results, calculate the overlap between the global path and the predicted trajectory of the dynamic obstacle, and determine whether local path reconstruction is needed by combining the preset overlap threshold.

[0096] It should be noted that the formula for calculating overlap is:

[0097]

[0098] In the formula, D represents the overlap between the global path and the predicted trajectory of dynamic obstacles; N represents the total number of dynamic obstacles; and k represents the index value. Indicates the end of time; δ represents the starting point of time; δ represents the indicator function; Path(t) represents the spatial position of the global path at time t; μ i (t) represents the spatiotemporal region occupied by the i-th dynamic obstacle at time t; t represents the time variable; P represents the empty set; occ (t) represents the spatiotemporal occupancy probability of a dynamic obstacle at time t; T total This represents the total effective time span of the predicted trajectory for dynamic obstacles.

[0099] S344. If the overlap is not less than the preset overlap threshold, local path reconstruction is triggered to generate a new local path, which is then spliced ​​with the global path to generate an obstacle avoidance path; otherwise, the global path is retained and an obstacle avoidance path is generated.

[0100] S35. Obtain multi-dimensional scene parameters through vehicle-to-everything (V2X) technology, and use fuzzy logic reasoning algorithm to map the multi-dimensional scene parameters into path optimization indicators. Adjust the spatiotemporal parameters of the obstacle avoidance path based on the path optimization indicators to generate an obstacle avoidance strategy.

[0101] It should be further explained that, based on vehicle-to-everything (V2X) technology (5G+V2X network communication technology), connecting vehicles and road-vehicle units, and coordinating vehicle-road collaboration in the connected environment perception stage, it supports vehicles to obtain more information than single-vehicle intelligent perception, such as non-line-of-sight perception or solving problems such as susceptibility to adverse environmental conditions; and coordinating in the computing and decision-making stage, it increases the systematic decision-making between vehicles and roads, such as solving vehicle priority management and traffic intersection optimization control.

[0102] S4. Construct a hierarchical simulation framework and establish a coupled model in the digital twin platform. Use model predictive control to conduct obstacle avoidance tests and optimize the obstacle avoidance strategy based on the simulation test results.

[0103] Preferably, a hierarchical simulation framework is constructed, and a coupled model is established in the digital twin platform. Obstacle avoidance testing is conducted using model predictive control, and the obstacle avoidance strategy is optimized based on the simulation test results, including:

[0104] S41. Based on the dynamic model of intelligent connected vehicles, and combining the static obstacle constraint diagram and the dynamic obstacle spatiotemporal occupancy diagram, a hierarchical simulation framework is constructed.

[0105] S42. In the digital twin simulation platform, based on the hierarchical simulation framework, construct a coupled model of intelligent connected vehicles and the environment, and define the control input of the coupled model.

[0106] S43. Based on the coupled model, the model predictive control algorithm is used to track the obstacle avoidance trajectory of intelligent connected vehicles, and simulation tests are carried out in several different scenarios.

[0107] Preferably, based on the coupled model, a model predictive control algorithm is used to track the obstacle avoidance trajectory of the intelligent connected vehicle, and simulation tests are conducted in several different scenarios, including:

[0108] S431. Based on the coupling model, set the initial state and target point sequence of the intelligent connected vehicle, and load different simulation scenarios;

[0109] S432. Based on the generated obstacle avoidance strategy, set the optimization objective and corresponding constraints, and use the model predictive control algorithm to calculate the optimal control input;

[0110] S433. Based on the optimal control input, execute obstacle avoidance strategies in simulation scenarios with different loads, perform obstacle avoidance trajectory tracking, and record performance data in each simulation process.

[0111] S44. By analyzing the simulation test results, the obstacle avoidance strategy is evaluated. Based on the evaluation results, the obstacle avoidance strategy is optimized using a reinforcement learning algorithm.

[0112] It should be added that one method for intelligent obstacle avoidance based on intelligent connected vehicles is to develop intelligent connected vehicles in limited areas and specific scenarios, and to develop cost-effective and feasible intelligent driving vehicles that are suitable for different application scenarios such as parks and factories.

[0113] like Figure 2 As shown, according to another embodiment of the present invention, an intelligent obstacle avoidance system based on intelligent connected vehicles is also provided, comprising:

[0114] Vehicle location determination module 1 is used to collect multimodal environmental data, construct a three-dimensional environmental map using point cloud fusion technology, and determine the location information of the intelligent connected vehicle in the three-dimensional environmental map through a multi-sensor fusion positioning algorithm;

[0115] The obstacle recognition module 2 is used to identify the spatial distribution of static obstacles and the motion characteristics of dynamic obstacles based on the constructed 3D environment map using deep learning algorithms, and to predict the motion trajectory of dynamic obstacles through neural networks.

[0116] The obstacle avoidance strategy generation module 3 is used to generate an obstacle avoidance path based on the location information of the intelligent connected vehicle, the spatial distribution of static obstacles and the predicted trajectory of dynamic obstacles, using a path planning algorithm, and to optimize the obstacle avoidance path by combining multi-dimensional scene parameters to obtain an obstacle avoidance strategy.

[0117] The obstacle avoidance strategy optimization module 4 is used to construct a hierarchical simulation framework and establish a coupled model in the digital twin platform. It uses model predictive control to conduct obstacle avoidance tests and optimizes the obstacle avoidance strategy based on the simulation test results.

[0118] In summary, by leveraging the technical solutions described above in this invention and integrating multimodal sensor fusion and point cloud mapping technologies, three-dimensional modeling of the surrounding environment can be achieved. Simultaneously, combined with multi-sensor localization algorithms, accurate vehicle location information can be obtained, providing a solid foundation for subsequent path planning and obstacle avoidance strategies. This significantly enhances the perception capabilities and reaction accuracy of intelligent connected vehicles in complex traffic environments. By combining deep learning-based obstacle recognition and trajectory prediction technologies, intelligent connected vehicles can pre-perceive and assess potential risks in dynamic environments. This allows vehicles to optimize path planning and obstacle avoidance strategies based on multi-dimensional environmental parameters, ensuring not only vehicle safety in complex and dynamic environments but also improving traffic efficiency and guaranteeing smooth and efficient driving under various road conditions. Furthermore, by utilizing a hierarchical simulation framework and a digital twin platform to conduct large-scale, multi-scenario obstacle avoidance strategy testing in a virtual environment, the adaptability of the obstacle avoidance system is enhanced, and the robustness of control is improved, enabling intelligent connected vehicles to operate stably and effectively avoid obstacles under various conditions.

[0119] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. 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 intelligent obstacle avoidance based on intelligent connected vehicles, characterized in that, include: S1. Collect multimodal environmental data, construct a three-dimensional environmental map using point cloud fusion technology, and determine the location information of the intelligent connected vehicle in the three-dimensional environmental map through a multi-sensor fusion positioning algorithm; S2. Based on the constructed 3D environment map, deep learning algorithms are used to identify the spatial distribution of static obstacles and the motion characteristics of dynamic obstacles, and the motion trajectory of dynamic obstacles is predicted through neural networks. S3. Based on the location information of the intelligent connected vehicle, the spatial distribution of static obstacles and the predicted trajectory of dynamic obstacles, an obstacle avoidance path is generated using a path planning algorithm. The obstacle avoidance path is then optimized by combining multi-dimensional scene parameters to obtain an obstacle avoidance strategy. S4. Construct a hierarchical simulation framework and establish a coupled model in the digital twin platform. Use model predictive control to conduct obstacle avoidance tests and optimize the obstacle avoidance strategy based on the simulation test results.

2. The intelligent obstacle avoidance method based on intelligent connected vehicles according to claim 1, characterized in that, The process of collecting multimodal environmental data, constructing a 3D environmental map using point cloud fusion technology, and determining the location information of the intelligent connected vehicle in the 3D environmental map using a multi-sensor fusion positioning algorithm includes: S11. Collect multimodal environmental data through several types of sensors, and perform noise reduction, correction and spatiotemporal alignment processing on the multimodal environmental data; S12. Extract point cloud features from the processed multimodal environment data, and use point cloud fusion technology to perform multi-frame registration and global optimization in order to construct a three-dimensional environment map. S13. Based on the initial pose information, Kalman filtering is used to fuse multimodal environmental data, and a multi-sensor fusion positioning algorithm is used to perform real-time map matching to determine the real-time location information of the intelligent connected vehicle in the three-dimensional environment map.

3. The intelligent obstacle avoidance method based on intelligent connected vehicles according to claim 1, characterized in that, The constructed 3D environment map utilizes deep learning algorithms to identify the spatial distribution of static obstacles and the motion characteristics of dynamic obstacles, and predicts the motion trajectory of dynamic obstacles through neural networks, including: S21. Based on the constructed 3D environment map, the point cloud data is preliminarily processed using the point cloud semantic segmentation algorithm in deep learning to distinguish different categories of static obstacles. S22. Use the Euclidean clustering algorithm to perform secondary processing on the preliminary processing results to generate static obstacle point cloud clusters containing semantic labels and obtain the spatial distribution of static obstacles. S23. Apply target detection algorithms to point cloud data to detect and classify dynamic obstacles, and combine spatiotemporal convolutional networks to extract the motion features of dynamic obstacles. S24. Construct a multi-task learning framework, combine the motion characteristics and time series data of dynamic obstacles, and use neural networks to predict the motion trajectory of dynamic obstacles.

4. The intelligent obstacle avoidance method based on intelligent connected vehicles according to claim 1, characterized in that, The obstacle avoidance strategy is obtained by generating an obstacle avoidance path using a path planning algorithm based on the location information of the intelligent connected vehicle, the spatial distribution of static obstacles, and the predicted trajectory of dynamic obstacles, and then optimizing the obstacle avoidance path by combining multi-dimensional scene parameters. S31. Based on the spatial distribution of static obstacles, obtain the expansion region of static obstacles and set the corresponding safe distance threshold. S32. Based on the expansion region and the safety distance threshold, the three-dimensional environment map is divided into different regions to construct a static obstacle constraint map; S33. Using the location information of the intelligent connected vehicle as the starting point and the target location as the ending point, combined with the static obstacle constraint diagram, the A* algorithm is used to perform path planning and generate a global path. S34. Using the predicted trajectory of dynamic obstacles as a local time-varying constraint, the global path is locally reconstructed using the fast expanding random tree algorithm to generate an obstacle avoidance path. S35. Obtain multi-dimensional scene parameters through vehicle-to-everything (V2X) technology, and use fuzzy logic reasoning algorithm to map the multi-dimensional scene parameters into path optimization indicators. Adjust the spatiotemporal parameters of the obstacle avoidance path based on the path optimization indicators to generate an obstacle avoidance strategy.

5. The intelligent obstacle avoidance method based on intelligent connected vehicles according to claim 4, characterized in that, The process of obtaining the expansion region of static obstacles based on their spatial distribution and setting corresponding safety distance thresholds includes: S311. Based on the spatial distribution of static obstacles, the point cloud data of static obstacles is reconstructed into a triangular mesh model, and the triangular mesh model is discretized. S312. Based on the discretization results, calculate the shortest distance from each triangular mesh vertex to the static obstacle, and establish a symbolic distance field; S313. Based on the established symbolic distance field, segmentation is performed using a preset distance threshold to define the area within a preset range of the static obstacle as the expansion region. S314. Set the corresponding safety distance threshold according to the expansion area and the specifications and dimensions of the intelligent connected vehicle.

6. The intelligent obstacle avoidance method based on intelligent connected vehicles according to claim 5, characterized in that, The different areas include: free areas, restricted areas, and restricted areas.

7. The intelligent obstacle avoidance method based on intelligent connected vehicles according to claim 6, characterized in that, The step of using the predicted trajectory of dynamic obstacles as a local time-varying constraint and employing a fast extended random tree algorithm to locally reconstruct the global path to generate an obstacle avoidance path includes: S341. Based on the predicted trajectory of the dynamic obstacle, the four-dimensional space is rasterized and the dynamic obstacle is assigned a spatiotemporal occupancy probability to construct a spatiotemporal occupancy map. S342. Combining the spatiotemporal occupancy graph with the fast expanding random tree algorithm, the global path is expanded, and the spatiotemporal occupancy probability of the newly expanded node to the global path is queried through collision detection. S343. Based on the collision detection results, calculate the overlap between the global path and the predicted trajectory of the dynamic obstacle, and determine whether local path reconstruction is needed by combining the preset overlap threshold. S344. If the overlap is not less than the preset overlap threshold, local path reconstruction is triggered to generate a new local path, which is then spliced ​​with the global path to generate an obstacle avoidance path; otherwise, the global path is retained and an obstacle avoidance path is generated.

8. The intelligent obstacle avoidance method based on intelligent connected vehicles according to claim 1, characterized in that, The construction of a hierarchical simulation framework, the establishment of a coupled model in a digital twin platform, the use of model predictive control for obstacle avoidance testing, and the optimization of obstacle avoidance strategies based on simulation test results include: S41. Based on the dynamic model of intelligent connected vehicles, and combining the static obstacle constraint diagram and the dynamic obstacle spatiotemporal occupancy diagram, construct a hierarchical simulation framework. S42. In the digital twin simulation platform, based on the hierarchical simulation framework, construct a coupled model of intelligent connected vehicles and the environment, and define the control input of the coupled model; S43. Based on the coupled model, the model predictive control algorithm is used to track the obstacle avoidance trajectory of intelligent connected vehicles, and simulation tests are carried out in several different scenarios. S44. By analyzing the simulation test results, the obstacle avoidance strategy is evaluated. Based on the evaluation results, the obstacle avoidance strategy is optimized using a reinforcement learning algorithm.

9. The intelligent obstacle avoidance method based on intelligent connected vehicles according to claim 8, characterized in that, The method of using a model predictive control algorithm based on a coupled model to track the obstacle avoidance trajectory of intelligent connected vehicles, and conducting simulation tests in several different scenarios, includes: S431. Based on the coupling model, set the initial state and target point sequence of the intelligent connected vehicle, and load different simulation scenarios; S432. Based on the generated obstacle avoidance strategy, set the optimization objective and corresponding constraints, and use the model predictive control algorithm to calculate the optimal control input; S433. Based on the optimal control input, execute obstacle avoidance strategies in simulation scenarios with different loads, perform obstacle avoidance trajectory tracking, and record performance data in each simulation process.

10. An intelligent obstacle avoidance system based on intelligent connected vehicles, used to implement the intelligent obstacle avoidance method based on intelligent connected vehicles as described in any one of claims 1-9, characterized in that, include: The vehicle location determination module is used to collect multimodal environmental data, construct a three-dimensional environmental map using point cloud fusion technology, and determine the location information of the intelligent connected vehicle in the three-dimensional environmental map through a multi-sensor fusion positioning algorithm. The obstacle recognition module is used to identify the spatial distribution of static obstacles and the motion characteristics of dynamic obstacles based on the constructed 3D environment map using deep learning algorithms, and to predict the motion trajectory of dynamic obstacles through neural networks. The obstacle avoidance strategy generation module is used to generate an obstacle avoidance path based on the location information of the intelligent connected vehicle, the spatial distribution of static obstacles and the predicted trajectory of dynamic obstacles, using a path planning algorithm. The obstacle avoidance path is then optimized by combining multi-dimensional scene parameters to obtain an obstacle avoidance strategy. The obstacle avoidance strategy optimization module is used to build a hierarchical simulation framework and establish a coupled model in the digital twin platform. It uses model predictive control to conduct obstacle avoidance tests and optimizes the obstacle avoidance strategy based on the simulation test results.

Images