3D Object Segmentation of Point Clouds using Profiling Techniques

Abstract

In the automatic processing of point clouds, higher level information in the form of point segments is required for classification and object detection purposes. Segmentation allows for the definition of these segments. Because of the increasing size of point clouds faster and more reliable segmentation methods are being sought. Various algorithms have been proposed for the segmentation of point clouds. In this paper, an extension of a segmentation approach based on intersecting profiles is proposed. In the presented method, surfaces are considered as a graph of intersecting planar curves. In this graph structure curves intersect at common points and terminate at surface discontinuities. This property of the curves makes it possible to determine point segments by connected components. A method for the detection of curves in the profiles is presented. The algorithm has been tested on terrestrial lidar point clouds.