Automatic Building Detection Using Airborne LIDAR Data

A method automatically extracting buildings from LIDAR data is presented. LIDAR point clouds are re-sampled into regular grid DSM, and a filtering processing which filters out non-ground points on the DSM surface is carried out simultaneously, acquiring a DEM with only ground points. An nDSM is obtained by subtracting DEM from DSM, and the building candidate regions are identified by using some thresholds to the nDSM. For these building candidates, we extract texture features in each candidate region, that is image texture based on Gray Level Co-occurrence Matrix (GLCM). Unsupervised clustering method is used with these features to separate building candidates from trees and vegetations. The workflow is presented in this paper and an example for a test site in Glenville approves it.