Research On Environmental Recognition And Local Path Planning Technology Based On Lidar

Road traffic safety problems caused by global traffic accidents have not been solved for a long time,so autonomous driving technology has received widespread attention as a major solution to traffic safety problems.Local path planning is one of the indispensable part in self-driving technology.Local path planning not only faces the problem about collision avoidance in road traffic,but also the comfort and efficiency of the vehicle.The detection of obstacles in the road environment is a prerequisite for local path planning to avoid collisions,so it is necessary to incorporate the detection of obstacles in the environment into the research of local path planning technology.The research is divided into two parts,including research of obstacle detection and local path planning based on lidar.The details are as follows:In the environment perception part,we use 3D lidar to collect environmental information,which obtains a wide range of point cloud data.The data processing process is mainly divided into obstacle detection and tracking.Firstly,in order to remove the background point cloud,the RANSAC and GPF algorithms were used to segment the road surface,and experiments proved that GPF has a more stable and accurate effect when processing ground data points.After this,the multi-clustering threshold-based Euclidean clustering algorithm was used to segment obstacles.In order to extract the motion information from the objects,a square box model of the obstacle is established based on the clustering results of the obstacle and the relevant features are extracted including geometric shapes,positions,and strengths.With these features,a multi-feature nearest neighbor association algorithm is proposed to complete the association matching of obstacles in the previous and subsequent frames.Finally we update the motion state of the obstacles by the Kalman filter algorithm.In the local path planning part of the unmanned vehicle(trajectory planning),this paper first converts the lower planning problem of the Cartesian coordinate system to the Frenet coordinate system for calculation,which reduces the computational requirements of the algorithm.In order to solve the three-dimensional trajectory,this paper decomposes it into the optimization problem of the horizontal and vertical trajectory in Frenet coordinate system.In the process of optimizing the horizontal and vertical trajectories,a combination of polynomials and sampling is used to fit the trajectory clusters,and then a trajectoryevaluation function is designed according to the requirements of comfort and driving efficiency to solve the optimal trajectory.The goals of trajectory optimization are different in different scenarios.In order to ensure that the algorithm can achieve better results,this article divides all scenarios in vertical planning into four working conditions,constant speed cruise,car following cruise,lane change convergence and Parking conditions.In order to solve the collision avoidance problem of static and dynamic obstacles,a secondary collision detection method is designed to detect the planned trajectory to ensure the safety of driving.In order to ensure the feasibility of the trajectory,the relevant constraints of the vehicle during the driving process are analyzed,including the steering system's mechanical constraints,the vehicle's kinematics,and dynamic constraints.Finally,the simulation scenario is built through the python visual programming library matplotlib for simulation.The simulation results show that the trajectory planning method proposed in this paper can complete the driving tasks under various conditions under the premise of ensuring safety,comfort and driving efficiency requirements.