The Study On SLAM Map Construction Algorithm Based On Lidar

Since its first proposal in 1988,Simultaneous localization and mapping(SLAM)have been considered by many scholars to be the key to the truly realization of a fully autonomous mobile robot due to its important theoretical and applied value.At present,it has been successfully applied to autonomous driving,unmanned patrol inspection,and geographic surveying,and other fields.In view of the high precision of 3D lidar and the advantages of being less susceptible to environmental factors,this thesis proposes an improved map construction scheme for mobile robots in unknown environments based on the analysis of the classic framework of SLAM systems.The main research contents are as follows:Firstly,the point cloud preprocessing operation is carried out on the data obtained by the threedimensional lidar.The ground information is extracted through the proposed ground point cloud segmentation model,which does not participate in the subsequent inter frame matching and loop detection,so as to reduce the amount of point cloud calculation.At the same time,the point cloud of the mobile robot’s own model and the point cloud of environmental noise are eliminated to avoid the interference of singular data to the algorithm.In order to improve the accuracy of the map,the matching method of integrating the normal distributions transform(NDT)algorithm and the iterative closest point(ICP)algorithm is used in the front-end matching.The NDT algorithm is used for rough matching to improve the initial state of point cloud and ensure that the ICP algorithm can get better convergence effect in fine matching.The fused algorithm not only improves the accuracy of point cloud matching,but also has certain robustness,and can be applied to a wider range.Secondly,the back-end optimization algorithm is further improved based on the loop detection theory.Aiming at the situation that loop detection can only optimize the trajectory error in the loop path and can not correct the pose information outside the loop path,a pose compensation scheme is proposed.The single frame drift correction value solved in the loop detection area is used as the compensation for the non loop path trajectory optimization.After pose compensation,the map accuracy is improved by about 3.3%.After the ground information extracted in the point cloud preprocessing stage is used to spatially constrain the 3D point cloud map,the average drift scale of the map constructed by the algorithm is reduced by 0.18 m,and the improvement is about 9.1%.Finally,the improved slam mapping algorithm proposed in this thesis is verified by KITTI public data set,and the operation results of the current mainstream open source algorithm Le GOLOAM are compared.All data indicators in the absolute trajectory error calculated by EVO evaluation tool have been greatly improved.The mapping effect in the actual scene can basically reproduce the details of the real environment and establish accurate map information.