OBJECT-BASED CLASSIFICATION OF URBAN AIRBORNE LIDAR POINT CLOUDS WITH MULTIPLE ECHOES USING SVM

Abstract. Airborne LiDAR point clouds classification is meaningful for various applications. In this paper, an object-based analysis method is proposed to classify the point clouds in urban areas. In the process of classification, outliers in the point clouds are first removed. Second, surface growing algorithm is employed to segment the point clouds into different clusters. The above point cloud segmentation is helpful to derive useful features such as average height, size/area, proportion of multiple echoes, slope/orientation, elevation difference, rectangularity, ratio of length to width, and compactness. At last, SVM-based classification is performed on the segmented point clouds with radial basis function as kernel. Two datasets with high point densities are employed to test the proposed method, and three classes are predefined. The results suggest that our method will produce the overall classification accuracy larger than 97% and the Kappa coefficient larger than 0.95.