Classification of 3D UAS-SfM Point Clouds in the Urban Environment

Abstract

The classification of three-dimensional (3D) point clouds derived through the use of cost-effective and time-efficient photogrammetric technologies can provide helpful information for applications, particularly in the mapping context. This paper presents a practical study of 3D Unmanned Aerial System (UAS) – Structure-from-Motion (SfM) point cloud classification using mainly open-source software. Following a supervised classification approach that makes use of only the dimensionality of points, the entire scene was classified into three land-cover categories: ground, high vegetation, and buildings. By applying the above-mentioned approach, the level of competence in classifying a 3D point cloud of a heterogeneous scene situated in the University of KwaZulu-Natal, South Africa, was evaluated. The resulting overall classification accuracy of 81.3%, with a Kappa coefficient of 0.70, was determined by means of a confusion matrix. The results achieved indicate the potential use of open-source software and 3D UAS-SfM point cloud classification in mapping and monitoring complex environments and in other applications that might arise.