Research On Dynamic SLAM Algorithm Based On Multi-sensor Fusion

Positioning and navigation is one of the important research contents of outdoor unmanned driving.Traditional outdoor unmanned positioning and navigation mainly rely on GPS and Beidou.With the update of technology,SLAM(Simultaneous Localization and Mapping)technology has become one of the new development directions for unmanned positioning and navigation.However,due to the presence of dynamic objects such as pedestrians and vehicles in the environment,as well as wheel slip,uneven ground,and drastic changes in lighting,in order to solve these problems,the paper adopts multiple sensors such as Li DAR,camera,and IMU,and studies multi-sensor fusion algorithms to improve the reliability and accuracy of positioning.The specific work is as follows:(1)In view of the bad environment such as high dynamic and no GPS signal area,by replacing the laser feature point extraction module in the LOAM series algorithm,a slope based ground wall feature extraction method is proposed,and the minimum re projection error function based on point to surface is constructed to filter out the feature points of small dynamic objects and improve the robustness of the laser SLAM algorithm;(2)To address the issue of the LVI-SAM project not implementing IMU and Lidar online calibration functions,a spline curve based IMU and Lidar external parameter online calibration algorithm is proposed to achieve online calibration of the camera,IMU,and Lidar sensors;(3)A deep learning method is proposed to detect dynamic regions in complex dynamic environments where moving vehicles,people,and other objects affect the accuracy of the SLAM algorithm.Then,a motion consistency detection algorithm is used to verify and remove the moving regions,and the remaining visual and laser feature points are extracted;During the visual loopback process,semantic feature descriptors are used to detect loopback;In laser loopback engineering,TEASER registration is used instead of ICP algorithm to detect loopback.Compared to the LVI-SAM algorithm,this algorithm incorporates online camera calibration,IMU,and Lidar external reference functions,improving the accuracy of pose calculation;(4)By building experimental equipment for data collection in industrial parks,the following algorithms are proposed: firstly,a laser SLAM based on wall and ground feature points is proposed,which greatly improves the feature extraction speed compared to open-source algorithms;Secondly,a joint calibration algorithm based on B-spline curves for Li DAR,camera,and IMU is proposed,which can provide online calibration values for three sensors during program operation compared to open-source algorithms;Thirdly,a dynamic SLAM algorithm based on multi-sensor fusion was proposed.By verifying the industrial park dataset,the output results basically coincide with the RTK results.Figure[40] table[4] reference[66]...