K-MEANS CLUSTERING BASED ON OMNIVARIANCE ATTRIBUTE FOR BUILDING DETECTION FROM AIRBORNE LIDAR DATA

Building detection is an important process in urban applications. In the last decades, 3D point clouds derived from airborne LiDAR have been widely explored. In this paper, we propose a building detection method based on K-means clustering and the omnivariance attribute derived from eigenvalues. The main contributions lie on the automatic detection without the need for training and optimal neighborhood definition for local attribute estimation. Additionally, one refinement step based on mathematical morphology (MM) operators to minimize the classification errors (commission and omission errors) is proposed. The experiments were conducted in three study areas. In general, the results indicated the potential of proposed method, presenting an average Fscore around 97%.