A Structure Sensitive Algorithm for Building Feature Line Extraction from LiDAR Data

Abstract

Lines are important features for Airborne Laser Scanning (ALS) point clouds processing and model reconstruction, in which the lines are often detected by a Hough Transformation (HT) similar to in image processing. In fact, feature lines represent the edges of buildings, which are man-made objects and often have orthogonal, parallel and other relationships of regularity. In this paper, we propose a method for detecting and refining the line features from ALS data in consideration of these relationships of regularity. First, an angle and voting algorithm is applied to conduct line detection to obtain the primary results. Second, an optimization process called structure sensitive competition, which relies on a line stability descriptor (LSD), refines the detected line segments. Finally, this proposed method is tested and compared to HT algorithm on a group of buildings with different complexity. The quality indicators, completeness, correctness and quality, show that the quality of the extracted lines can be substantially improved after the structure sensitive optimization.