Fig. 1. Flow chart of the proposed algorithm

Fig. 2. Extract ground features point from LiDAR cloud and ROI

Fig. 3. Feature correspondences after CRI verify

Fig. 4. Process of feature point correspondence searching in Lidar odometry

Fig. 5. Comparison of feature point correspondence process. (a) Search the correspondence directly; (b) local details of direct search process; (c) search correspondence after transformed by camera motion a priori; (d) local details of search process after transform

Fig. 6. Comparison of trajectory and the real scene. (a) Comparison of real trajectory and trajectories estimated by LeGO-LOAM-IMU, LeGO-LOAM-noIMU, and proposed, respectively; (b) satellite map of the trajectory

Fig. 7. Comparison of the estimated value from proposed algorithm, LeGO-LOAM-IMU, LeGO-LOAM-noIMU and the real value on x,y, and z axis

Fig. 8. Robot and laboratory environment. (a) Robot; (b) laboratory environment

Fig. 9. Estimated trajectory of laboratory environment

Fig. 10. Experimental results of outdoor environment. (a) Comparison of trajectory estimated by algorithms; (b) point cloud map; (c) satellite map