A progressive approach for building extraction by fusing LIDAR data and co-registered bands

In this paper, a progressive approach is presented for building extraction from LIDAR combined with its co-registered bands. Carefully tuned Gabor wavelet filters are applied to LIDAR data for object detection on the earth, and to localize targeted objects in order to minimize hilly terrain effects. Object classification is then carried out in local areas where objects exist. The Dempster-Shafer (DS) theory of evidence is used to conduct initial classification by fusing LIDAR and its co-registered data sets; and a fuzzy Markov random field (FMRF) model takes the output from the DS theory of evidence, and is performed the further object classification to extract buildings. Finally the level set approach is employed for building boundary extraction. The testing experiments under this research have shown the potential of this approach in accurately extracting buildings from airborne LIDAR and its co-registered bands.