Edge and tree detection from three-dimensional unorganised point clouds from terretrial laser scanners

Abstract

A method for edge detection in unorganised point clouds from terrestrial laser scanners without preprocessed triangle meshes is proposed. The method is based on geometric primitives such as change of geometric curvature and surface normal vector. As an application of the proposed edge detection, we introduce an algorithm for tree detection. Using the fact that trees have higher curvature points than human-made structures, e.g. buildings, they can be extracted from the point cloud. Higher curvature points are used as seeds for a region-growing method to detect whole trees with the aid of a reasonable termination condition.