Morphology-based building detection from airborne lidar data

The advent of Light Detection and Ranging (LIDAR) technique provides a promising resource for three-dimensional building detection. Most current methods commonly fuse LIDAR data with other multi-spectral images to help remove vegetation based on NDVI or other vegetation indices; however, the fusing process may cause errors that are introduced by resolution differences, geo-referencing, time differences, shadow and high-rise building displacement problems. Due to the difficulty of removing vegetation, relatively few approaches have been developed to detect buildings only from LIDAR data. This paper presents a morphological building detection method to identify buildings by gradually removing non-building pixels. First, a ground filtering algorithm separates ground from buildings, trees, and other objects. Then an analytical approach further removes the remaining non-building pixels using size, shape, height, building element structure, and height difference between the first and last return. The experiment results show this method provides a comparative performance with an overall accuracy of 95.46% as in the study site in the Austin urban area.