A Robust Model Reconstruction Algorithm For Elevator Shaft

Abstract

In this paper we introduce an elevator shaft model reconstruction algorithm based on low-cost vehicle LiDAR and IMU. To obtain the accurate odometry of LiDAR, we fuse loop closure detection module based on Scan Context descriptor with LiDAR-Inertial odometry and reconstruct the map at the same time. In the front-end, after recieving LiDAR and IMU data, iterated error state Kalman filter is used to optimize the robot pose and map which are then stored in an iVox map. In the back-end, we extract Scan Context descriptors of each LiDAR scan and compare with history scans to detect loop closure and further correct robot pose and map. Finally the pose and map in front-end are updated by results of back-end. To carry out the experiment in elevator shaft, we design our own working platform including scanning equipment and unwinding equipment. Our algorithm shows high robustness and accuracy in both private dataset in elevator shaft scenario and public dataset in other environments.