Delay-CLD Algorithm For Closed-loop Detection And Its Application In 3D Laser SLAM

Since SLAM(Simultaneous Localization and Mapping)was proposed in 1988,after years of research and development,it gradually showed great application value in high-tech fields such as artificial intelligence and unmanned driving.Due to the errors of the related sensors used in SLAM and the flaws of the SLAM algorithm itself,SLAM will inevitably have mapping and positioning drift during operation.In this regard,an additional algorithm is needed to provide optimization instructions for the mapping and positioning process of SLAM.This algorithm is usually implemented in SLAM by detecting whether it passes through the same location.It is a closed-loop detection algorithm.In academia,this method is called a closed-loop detection algorithm.The 3D laser SLAM and the closed-loop detection algorithm for 3D laser SLAM have been research hotspots in the field of SLAM in recent years.After studying the global development status of 3D laser SLAM and related closed-loop detection methods,this paper has made the following preparations for the research of closed-loop detection for 3D point cloud data: First,this paper has studied the working principle of lidar.On this basis,the mechanical rotary multi-line lidar is selected as the main sensor for this study;secondly,the overall framework of laser SLAM is introduced to clarify the position of closed-loop detection in the framework;finally,several commonly used point cloud feature extraction methods are compared and determined the research direction of closed-loop detection based on point cloud feature descriptors.In order to efficiently extract robust feature descriptors from 3D point cloud data,this paper proposes an optimization algorithm Fast-M2 DP based on the original 3D global descriptor M2 DP algorithm: inherits the idea of M2 DP algorithm to project a 3D point cloud onto a 2D plane To extract feature descriptors,and optimize the serial realization node of M2 DP with CPU / GPU parallel method.Experimental results show that Fast-M2 DP not only ensures the richness of feature descriptor information,but also improves the efficiency of the algorithm.In order to improve the running speed and accuracy of closed-loop detection,this paper proposes a complete closed-loop detection algorithm framework Delay-CLD:(1)Set a delay detection threshold based on laser SLAM front-end pose estimation to avoid a lot of meaningless calculations.Fast-M2 DP is used to determine closed-loop candidates efficiently and robustly after the delay detection conditions are met;(2)Super G-4PCS-based coarse registration is performed between the determined closed-loop candidates to ensure that erroneous closed-loop detection is not handled,And then use the improved ICP algorithm to perfect the conversion estimation,and finally complete the closed-loop detection.Experiments show that the proposed closed-loop detection algorithm has good performance.Finally,this paper uses Delay-CLD on the ROS platform to transform the classic 3D laser SLAM algorithm LOAM,and performs the real map construction and positioning of the algorithm before and after the transformation in atypical indoor scenes.Figure[63] Table[18] Reference[71]...