Color-based Segmentation of Point Clouds

Segmentation is one of the most fundamental procedures for the automation of point cloud processing. The methods based on geometrical derivatives such as curvature and normals often lead to over-segmentation and even failure when used to segment point clouds of geometrically-complex architectures. In this paper we present a point cloud segmentation algorithm based on colorimetrical similarity and spatial proximity. The algorithm contains region growing, region merging and refinement processes. The region growing process uses kd-tree to search the k-nearest neighbors of each seed point. The resulting regions are then merged and finally refined on the basis of colorimetrical and spatial relation. In each of the process, we developed different criteria corresponding to the different tasks to carry on the segmentation. The algorithm requires a small number of manually set parameters which are used to keep balance between underand over-segmentation. The experiments of the presented algorithm on a point cloud of Chinese ancient architecture show its effectiveness. The segmented regions can be used to reconstruct 3D models of different parts of the architectures.