Chuanjiang Li, Ziwei Hu, Yanfei Zhu, Xingzhao Ji, Chongming Zhang, and Ziming Qi
SLAM, location, keyframe, mapping
This paper presents lightweight and fast Lidar-inertial odometry (LF-LIO) for a robot’s real-time pose estimation in an unknown complex environment. This system includes prediction, odometry, mapping, and trajectory optimisation modules. In the prediction module, the initial value of the odometer’s motion is calculated by inertial measurement unit (IMU) pre-integration and the state of the previous moment, the odometry then employs a scan-to-submap matching method based on ground segmentation and optimisation proposed by this paper to estimate the pose transformation between consecutive frames. To ensure high performance in real-time, a keyframe map is created instead of a full map. When updating incrementally the efficiency of the map is improved, meanwhile an efficient dynamic sliding window is proposed to manage sub- maps. We compare the performance of LF-LIO with three methods, Lidar odometry and mapping in real-time (LOAM), lightweight and ground-optimised Lidar odometry and mapping on variable terrain (LeGO-LOAM), and fast direct Lidar-inertial odometry (Fast-LIO2), using KITTI datasets, the contrasted results of the application indicate that the proposed LF-LIO method has better accuracy with a reduced computational burden.
Important Links:
Go Back
