Efficiently Extracting Dominant Planes from Massive 3D Points Based on Scene Structure Priors

Extracting reliable dominant planes from 3D points plays an important role in modeling large-scale urban scenes. However, traditional local and global methods frequently appear powerless when massive noisy 3D points are present. In order to solve the problem, the paper proposes an efficient multi-plane extraction method to explore potential plane relations by detecting 2D line segments based on scene structure priors in a low-dimension projection map derived from 3D points, and then generate planes from the resulting plane relations. Experimental results confirm that our method can efficiently produce sufficient and reliable dominant planes from massive 3D points with high noise levels (only about 6s on 2000K 3D points).