Fig. 1. (a) 3D kd-tree (interior nodes are labeled as their splitting planes and leaf nodes are labeled in their boxes); (b) graph representation of the same kd-tree

Fig. 2. Tension. (a) Tension at the non-edge points; (b) tension at the edge points

Fig. 3. Case of selecting the new seed point when the point clouds are evenly aligned

Fig. 4. Search angle

Fig. 5. P2 will be ignored if the search angle is too small

Fig. 6. (a) Result of the space plane with sharp edge formed by the ideal point clouds; (b) extraction result at the sharp edge; (c) smooth effect at the depression edge obtained by the algorithm

Fig. 7. (a) Closed edge extracted by one initial seed edge point; (b) all edges extracted by six initial seed edge points; (c) effect of uneven distribution of point clouds on edge extraction

Fig. 8. Results for different neighbor radii and search angles of two intersecting cylinders. (a) R=R0, θ=100°; (b) R=R0, θ=135°; (c) R=1.5R0, θ=100°; (d) R=1.5R0, θ=135°

Fig. 9. Edge coefficient method

Fig. 10. Eight-neighbor depth method