In this study, we suggested flatness as new geometric feature to base on which to classify point cloud using a machine learning. In order to verify that usefulness, tree point clouds which was obtained by TLS were classified into leaf and wood class on a voxel basis. As classification method, K-means clustering based on flatness and was applied. Two tree species was target for classification. The result showed, thin branches could not be classified into wood class but tree species were not influenced to classification accuracy. Best OA and Kappa were 70.0 % and 40.0 % respectably. We expected that further test using a 3 dimensional line fitting to detect thin branches with improve the accuracy. Moreover, this method can apply to point cloud data obtained by Structure from Motion technique.
展开▼
