3D Point Cloud Classification of Natural Environments Using Airborne Laser Scanning Data

摘要

Airborne Laser Scanning technology has been recognized as a valuable and high precision topographic source for mapping and 3D modelling of the Earth surface. 3D point cloud classification in natural environments has become an important topic due to an increasing number o f applications in geomorphology, 3D cartography, 3D modelling and environmental issues. This article presents two different three dimensional classification approaches applied for Airborne Laser Scanning point clouds. Firstly, a supervised machine learning classification algorithm using random forests, decision trees and training datasets and secondly a semi-supervized algorithm composed by a cascade of binary classifiers, based on Support Vector Machines are used for labeling each LiDAR return of the input point cloud as one of the following categories: ground, water, vegetation and gravel. The 3D classification results are demonstrated for a near natural reach of the Pielach River, the "Neubacher Au" area, located in Lower Austria, with Airborne Laser Scan ning data acquired from 26 February 2015, consisting in2.8 million LiDAR returns over 45150 m 2 , with a point density of 21 points/m 2 .