Fig. 1. Flow chart of the proposed algorithm

Fig. 2. Frame of global map and local graph optimization

Fig. 3. Platform of experiment

Fig. 4. Examples of two frame registration. (a) Point cloud before registration; (b) point cloud after registration

Fig. 5. Analysis of extended Kalman filter fusion location accuracy. (a) Error of x-axis; (b) error of y-axis; (c) error of z-axis

Fig. 6. Comparison of three different algorithms in indoor scene. (a) Results from 3D-NDT algorithm; (b) results from BLAM SLAM; (c) results from LOAM SLAM

Fig. 7. Comparison of three different algorithms in outdoor scene. (a) Results from 3D-NDT algorithm; (b) results from BLAM SLAM; (c) results from LOAM SLAM

Fig. 8. Effect of 3D-NDT mapping. (a) Real indoor scene; (b) point cloud map of indoor scene; (c) point cloud map of outdoor scene; (d) real outdoor scene