A graph-based plane segmentation approach for noisy point clouds

Semantic mapping is a long term goal to understand environment for mobile robots. The indoor environments usually consist of a large amount of planar surfaces. Thus, plane segmentation is an essential prerequisite to build a semantic map. In this paper, we develop an algorithm to segment planar surfaces from noisy point clouds of indoor scenes. The proposed segmentation algorithm is based on a graph-based representation of the 3D data to determine the mergence of two adjacent regions, which is able to detect all the planes accurately. We apply the algorithm to plane segmentation and illustrate the results with the synthetic point cloud and two point clouds of real-world indoor scenes, respectively. The experiment results show that our proposed segmentation algorithm is accurate to extract all the planar surfaces and adaptive to cope with the point cloud noise.