Research On SLAM Algorithm Of Lidar And Vision Fusion

Simultaneous localization and mapping is the key technology for robots to achieve full autonomy and multi-function.Mobile robots are mainly equipped with different sensory environment information sensors to achieve SLAM.However,a single sensor has significant limitations in implementing SLAM.In order to solve the problem which uses sensor information fusion strategy to achieve SLAM in this thesis.This thesis mainly uses lidar and depth cameras as sensors to sense the environment.It uses lidar to sense the two-dimensional information of the environment.It uses depth cameras to obtain color images and depth image information.The three-dimensional environment information obtained by the depth camera is projected onto the two-dimensional grid plane and the two-dimensional grid map obtained by the lidar is fused by the Bayesian criterion,to achieve improve the accuracy of building the map.Firstly,mathematical modeling of mobile robot,lidar and depth camera is carried out,and it analyze the map construction method and robot operating system in SLAM.Secondly,the laser SLAM algorithm is studied,and the traditional EKF SLAM,RBPF-SLAM and FastSLAM are studied and simulated.In view of the problem of the particle barrenness of RBPF-SLAM,an improved RBPF-SLAM algorithm is proposed in this thesis,and compared with the traditional algorithm through the simulation.The experimental results show that the improved algorithm is better,and the experiment of mapping in the actual environment also shows that the improved algorithm is more accurate in building maps.Thirdly,the RGB-D SLAM based on feature matching is studied,the traditional RGB-D SLAM is improved by the feature point extraction using the FLANN algorithm that can effectively shorten the time required for feature extraction,and the random sample consensus algorithm is used to removal of mismatched points in feature point extraction that can improve the matching accuracy.For camera motion estimation,the Bundle Adjustment algorithm based on minimizing reprojection errors is used to optimize the P3 P method,which greatly improves the accuracy of the trajectory estimation of camera motion.The fr1_desk,fr1_desk2,fr1_xyz in the TUM dataset are selected for verification for the running time after the improvement is compared with the running time before the improvement.The algorithm in this thesis is compared with the RMSE of RGB-D SLAM.The comparison shows that the algorithm in this thesis is better.Finally,in view of the problems that laser SLAM has in actual map construction,a SLAM algorithm for fusion of lidar and vision is proposed in this thesis.By two-dimensional projection of the three-dimensional map constructed by visual SLAM,according to the proposed fusion rules to achieve local integration of the constructed two-dimensional map.In order to test the effect of the fusion algorithm mentioned in thisthesis,an experimental platform and SLAM system for the fusion of lidar and depth camera were built.Through the experiment of the experimental platform in a real environment,the results show that the fusion algorithm can improve the existence of laser SLAM in a small-scale environment defect.