The Research On Object Detection And Path Planning For Vehicle Active Collision Avoidance

As traffic accidents and congestion problems become increasingly severe,the use of intelligent driving technology to achieve active collision avoidance through autonomous lane changing can be an effective way to avoid rear-end accidents and relieve traffic pressure.Environment perception and path planning are key technologies for achieving active collision avoidance.As the prerequisite for intelligent driving systems,environment perception requires to identify and locate obstacles ahead on the road accurately and in real time,which provides the necessary information for decision making and planning modules.Based on perception information and driving purpose,path planning requires to plan a drivable path in real-timet that meets vehicle and road constraints,and guides the vehicle to drive safely after the system has made a decision.Therefore,the accuracy of the perception and planning module directly affects the system’s ability to achieve the demand for active collision avoidance.This paper focuses on active collision avoidance demand,and conducts research on road object detection and path planning algorithms based on camera and Li DAR fusion.The main research contents are as follows:(1)To address the limited information of a single sensor,camera and Li DAR are used separately to achieve road object detection.For image object detection,the YOLOv7-tiny network model is utilized with improvements made to its performance through data set preprocessing and model parameter modification.After training,the m AP50 of the model is 80.1% and the FPS can reach 70.For point cloud object detection,a segment fitting method based on pre-selected seed points is used to segment ground point clouds,improving its adaptability to different road surfaces.The traditional DBSCAN algorithm has been improved by combining depth maps,improving the accuracy of clustering and improving real-time performance by 51.05%.Finally,feature suppression is performed based on the shape of the road object to return the desired object bounding box.(2)To address the problem of cumbersome processes and large errors for camera and lidar calibration,the camera and Li DAR external parameters are optimized and solved based on the co-planar constraints.So that the detection results are projected onto the same coordinate system for matching and fusion.By extracting the centroids and normal vector of the calibration board in the camera and Li DAR coordinate systems,establishing the coplanar constraints and setting the objective functions,the conversion matrix of Li DAR to camera coordinate system is obtained by the NSGA-II algorithm to achieve spatial alignment.At the same time,the soft synchronization of sensor data is realized based on the ROS timestamp.On this basis,the IOU method is used to pair and fuse the image and point cloud detection frame in the pixel coordinate system to obtain the complete object information.(3)To achieve collision avoidance path planning in road scenarios,an artificial potential field applicable to vehicle active collision avoidance is designed by trigonometric and exponential functions.The path planning results under various working conditions are analyzed,and it is found that the improved potential field overcomes the defects of the traditional method.It can constrain the path according to road parameters and adaptively adjust the potential field parameters by the relative speed between the vehicle and the obstacles.And the proposed path planning algorithm takes into consideration the kinematic limitations of the vehicle while ensuring smooth and comfortable driving,and maintaining safety constraints.The path is tracked in the Car Sim simulation environment and proved to have an excellent trackable performance.(4)To verify the feasibility of the algorithm in practical applications,a modified electric vehicle equipped with sensors such as camera,Li DAR,and GNSS/INS is used as the complete vehicle platform.An active collision avoidance scenario within the real campus scene is designed for experimental verification.The experimental results show that the optimized external parameter calibration method can accurately solve the conversion matrix of camera and LIDAR coordinate system,which projects the point cloud to the pixel plane better.The improved object detection and path planning algorithm can accurately identify and localize obstacles ahead,plan a safe and feasible vehicle active collision avoidance path.