Simultaneous Localization And Mapping Of Closed Park Based On Vehicle-borne LiDAR

Simultaneous Localization and Mapping(SLAM)technology plays an important role in unmanned driving.It localizes the unmanned vehicle and constructs the environmental map,which is crucial for unmanned vehicle to drive safely in unknown environment and complete specified tasks.LiDAR is an ideal sensor for constructing environmental map because of its stable performance,robustness to environment changes and accurate distance information.This paper systematically expounds the data processing of LiDAR,the front-end motion estimation of SLAM,the back-end motion optimization of SLAM and the construction of environmental map of the closed park area.The main contents of this paper include:(1)The data structure of LiDAR is analyzed,and a mass of point cloud data is stored and invoked in a way of lines.(2)The traditional Iterative Closest Point(ICP)is improved.Constraints of the point-to-line and the point-to-surface are established respectively according to the smoothness feature points and the surface feature points of point set.The Levenberg-Marquardt method is used for inter-frame motion estimation.(3)The point cloud map is divided into tridimensional grid,and then a part of the grid point cloud is selected to match the current frame point cloud.In the process of establishing constraints,a covariance matrix composed of point sets is constructed to improve the accuracy of feature point matching.(4)Octree three-dimensional raster map is constructed through point cloud map,which can achieve a better interaction and provide a basis for the following navigation and path planning.(5)Quantitative evaluation of mapping performance is made according to different sizes of tridimensional grid and covariance matrices with different amounts of point sets.The octree maps are built for two different closed parks using the best mapping scheme.The research results show that it is feasible to use SLAM technology to map closed parks,and the results are good.The method of extracting feature points by smoothness size is more efficient and accurate.The weakness of traditional iterative closest method is improved by using point-to-line and point-to-surface constraints,and the accuracy of inter-frame matching is improved.The results of interaction between high frequency motion estimation and low frequency motion optimization are very good,and the final map is well established.The size of the tridimensional grid of point cloud map has a certain impact on the mapping results,so the appropriate mesh size can make the mapping results better.The covariance matrix of different point sets has a significant impact on the mapping results,too many or too few point sets will increase the risk of mismatching between frames,appropriate number of point sets can make the optimization results better.The method of this paper can well realize the establishment of tridimensional grid map of closed park environment,meanwhile it realizes better human-machine interaction while realizing better map model.