Lightweight LiDAR-IMU odometry based on improved Kalman filter
LCG-LIO was proposed based on the FAST-LIO2 aiming at the problem of real-time localization and running poor stability of mobile robot. LCG-LIO exhibited lower computational complexity and increased localization accuracy compared with FAST-LIO2. The LCG-LIO frontend incorporated a method for extracting and segmenting high-quality plane and ground points by using the proposed bidirectional dimensionality reduction curvature filter in contrast to FAST-LIO2. LCG-LIO balances the number of plane and ground points through the pseudo occupancy of point clouds. Improvements were made to the observation error equations and the construction of its Jacobian matrix for the Kalman filter in the backend optimization. The GPS constraint was incorporated to the observation error equation by pseudo-trajectory weighting method,and cumulative odometry drift was corrected. Experimental validation was performed by using the KITTI dataset and self-collected datasets. Results showed that the accuracy and efficiency of the proposed method were improved by 55.13％ and 53.01％ compared with FAST-LIO2. The proposed method for integrating GPS has higher feasibility than the factor graph optimization in LIO-SAM.
万方数据
