SEGMENTATION AND CROWN PARAMETER EXTRACTION OF INDIVIDUAL TREES IN AN AIRBORNE TOMOSAR POINT CLOUD

Abstract. The analysis of individual trees is an important field of research in the forest remote sensing community. While the current state-of-theart mostly focuses on the exploitation of optical imagery and airborne LiDAR data, modern SAR sensors have not yet met the interest of the research community in that regard. This paper describes how several critical parameters of individual deciduous trees can be extraced from airborne multi-aspect TomoSAR point clouds: First, the point cloud is segmented by unsupervised mean shift clustering. Then ellipsoid models are fitted to the points of each cluster. Finally, from these 3D ellipsoids the geometrical tree parameters location, height and crown radius are extracted. Evaluation with respect to a manually derived reference dataset prove that almost 86% of all trees are localized, thus providing a promising perspective for further research towards individual tree recognition from SAR data.