Research On SLAM Algorithm Of 3D Laser Point Cloud Based On Mobile Robot

One of the current research hotspots in the field of intelligent robots is to enable mobile robots to realize positioning and mapping.Simultaneous localization and mapping(SLAM)technology is an important research direction to solve this problem.Li DAR SLAM is the most mature and stable technology until now,and the traditional multi-line Li DAR is difficult to be used in business widely due to its high cost,so this paper meanly focus on the study of 3D Li DAR SLAM using the Livox 3D solid-state Li DAR which has a lower price.The main contributions of this paper are as follows:(1)Aiming at the problem that the Livox solid-state Li DAR generates a large number of point clouds and unreliable noise points,we preprocess the point clouds data in this paper.Firstly,the bad points of the point clouds are eliminated,including the points that incident angle is too small,the reflection intensity change is too large,and the points hidden behind an objects,etc.,the confidence of these bad points is too low,which will affect the accuracy of the system.Then,the dense point cloud is down-sampled through the voxel method,which greatly reduces the number of point clouds while ensuring the preservation of environmental details,and it ensure the real-time performance of the algorithm.Finally,aiming at the point cloud scanning features of Livox solid-state Li DAR,a threshold segmentation algorithm was proposed,which eliminates unreliable points at a distance,and can further reduce the number of point cloud and improve the real-time performance of the algorithm.In order to verify the effectiveness of the point cloud preprocessing algorithm,the experiment compared the time and accuracy of the dense and sparse point cloud registration using the ICP algorithm before and after preprocessing.The results show that the preprocessed point cloud is better than the unpreprocessed point cloud.The accuracy difference is small,but the registration speed is significantly improved.(2)This paper proposes a feature extraction algorithm for Livox solid-state Li DAR point cloud data—Spherical Region Feature Histogram(SRFH).The SRFH feature extraction algorithm integrates the surrounding point clouds to construct a spherical region.A coordinate system is established in the spherical area,and a 3d histogram is established by using the point cloud direction and norm information in the area under this coordinate system,which is used as a feature for subsequent point cloud registration.(3)In point cloud registration,the point cloud registration algorithm based on Iterative Closest Point(ICP)is more sensitive to the selection of initial values.To solve this problem,first use SRFH feature-based Sample Consensus Initial Alignment(SAC-IA)algorithm performs coarse registration on the point cloud to obtain a coarse pose,and then uses this coarse pose as the initial value of the ICP algorithm to perform iterative calculations to obtain an accurate pose.Through experiments,we compare the registration time and accuracy of the dense and sparse point clouds based on the SRFH feature registration algorithm and the Fast Point Feature Histograms(FPFH)feature registration algorithm.The results show that the registration based on SRFH features Although the accuracy of the algorithm is lower than that of the registration algorithm based on FPFH features,it is more than 20 times faster than the algorithm and can meet the real-time requirements,which verifies the feasibility of the registration algorithm based on SRFH features.(4)Aiming at the cumulative error problem in mapping,we use loop detection to optimize the poses,extracting line cells and plane cells from key-frames to form a map,and express it in a 2D histogram.When the robot reach the location it have visited,the loop will be detected through the cell map,and finally the pose of the global map will be optimized through the pose graph optimization.Through simulation the possibility of using low-cost Livox solid-state Li DAR to realize Simultaneous localization and mapping on mobile robots is verified.And compared to the LOAM algorithm using expensive multi-line Li DAR,it has certain advantages in mapping accuracy.