Local Map Reconstruction And Road Edge Detection For Unmanned Ground Vehicle Based On A 3D Lidar In Cross-country Environment

The environment perception of the unmanned ground vehicle is the prerequisite to ensure the correct path planning.Compared with the structed environment,obstacle in crosscrountry environment is complicated and out of door.Moreover,bumpy road also brings more challenges to the unmanned ground vehicle environment perception.Among the sensors used in the unmanned ground vehicle,the 3D lidar has the advantages of high precision,large measurement range and strong robustness.It plays a more and more important role in the environment perception of the unmanned ground vehicle.Therefor,this paper mainly studies the key problem in the process of lidar detect in the cross-country environment.We choose local elevation map to describe complex bumpy cross-country environment.The vehicle motion is estimated used the data from integrated navigation system and vehicle information based on extended Kalman filter.After then,the 3D laser point cloud data is converted to the local horizontal coordinate system used the filter result to eliminate the disturbance of the vehicle bump.Using grid iterative closet point algorithm to realise the elevation map accurate registration,and the mean filter in the time domain to complete the local elevation map update.A real-time road boundary extraction algorithm based on road direction detection and two curve fitting is proposed.Using the grid beam model to statistic passing distance of the obstacle map in all directions,ray direction which passing distance exceeds the threshold value will do the neighborhood cluster,each cluster is a road direction,and take the average of all the ray angle inside the cluster as the road direction angle.Find the nearest obstacle point respectively as the seed point on both sides of the road boundary in the grid map along the direction of road.Curve fitting around the road boundary is then done using RANSAC algorithm.Finally particle filter algorithm is used for road boundary update in order to improve the stability of the algorithm.A large number of tests were done in the real environment with the unmanned vehicle as the "BIT IV No.1",finally,this paper summarizes the main work and contribution,and the next research work is prospected.