Automatic 3D Feature Extraction from Structuralized LIDAR Data

Abstract: LIDAR, or laser scanning, is capable of collecting accurate 3D coordinates of scanned points densely and sub-randomly distributed on scanned object surfaces. The huge amount of 3D points implies abundant recessive spatial information which can be turned into dominant information through various data processing methods. To explore valuable spatial information from LIDAR data automatically becomes an active research topic, for example extracting digital elevation model, buildings, and trees from LIDAR data. It has long been recognized that extracting features from implicit data is the first and essential step of deriving explicit information from data. In contrast to 2D features can be extracted from image data, this paper focuses on extracting 3D features from a point cloud data set. Because the most prominent features in point cloud are co-plane points, the proposed method begins with extracting 3D plane features. Then, 3D edges and corners can be extracted by intersecting neighboring planes. Most significant 3D features can be extracted automatically through the proposed data processing method. In order to handle the large amount of sub-randomly distributed point cloud data efficiently, organizing the data set is required during the data processing. This paper proposes an octree-based split-merge-intersect method to organize LIDAR point cloud and extract 3D features. The proposed method was applied on both airborne and ground LIDAR data. The test results show the promising capability of extracting 3D features from point cloud data. The need of economic computation time demonstrates the efficiency of the developed method.