Fig. 1. Normal vector of point cloud

Fig. 2. Normal vector of feature region and non-feature region. (a) Feature area; (b) non-feature area

Fig. 3. Feature point and non-feature point

Fig. 4. Algorithm for adaptive k-neighbor selection

Fig. 5. star model and local feature parameter distribution of the model. (a) star model; (b) LEFD distribution

Fig. 6. fandisk model and local feature parameter distribution map of the model. (a) fandisk model; (b) LEFD distribution

Fig. 7. Algorithm for edge detection

Fig. 8. Model point cloud visualization

Fig. 9. Comparison of results of complex models. (a)-(c) Original point cloud; (d)-(f) results of BE method;(g)-(i) results of the method in literature[10]; (j)-(l) results of the method in literature[10]; (m)-(o) results of our method

Fig. 10. Comparison of results of simple models. (a)-(c) Original point cloud; (d)-(f) results of BE method;(g)-(i) results of the method in literature[10]; (j)-(l) results of the method in literature[10]; (m)-(o) results of our method

Fig. 11. Dihedral angle with different angles. (a) 22°; (b) 45°; (c) 90°

Fig. 12. Results of noise immunity. (a) fandisk model, adding noise with standard deviation of 0.5%; (b) fandisk model, adding noise with standard deviation of 2%; (c) bunny model, adding noise with standard deviation of 0.5%; (d) bunny model, adding noise with standard deviation of 2%

Fig. 13. Data acquired by Kinect. (a) chair 1; (b) chair 2

Fig. 14. Detection of Kinect data with our method. (a) chair 1; (b) chair 2