Machine Learning-Based Supervised Classification of Point Clouds Using Multiscale Geometric Features

3D scene classification has become an important research field in photogrammetry, remote sensing, computer vision and robotics with the widespread usage of point clouds. Point cloud classification, called semantic labeling, segmentation, or clouds is a challenging topic. Machine learning, on other hand, powerful mathematical tool used to classify whose content can be significantly complex. In this study, performance different machine learning algorithms multiple scales was evaluated. The feature spaces points were created using geometric features generated based eigenvalues covariance matrix. Eight supervised tested four areas from three datasets (the Dublin City dataset, Vaihingen dataset Oakland3D dataset). evaluated terms overall accuracy, precision, recall, F1 score process time. best results obtained for test algorithms. Area 1 Random Forest as 93.12%, 2 Multilayer Perceptron algorithm 92.78%, 79.71% Support Vector Machines Linear Discriminant Analysis 97.30%.