Fusing LiDAR Data and Aerial Imagery for Building Detection Using a Vegetation-Mask-Based Connected Filter

Building detection is valuable for 3-D building reconstruction and urban management. In this letter, a vegetation mask-based connected filter (VMCF) algorithm is proposed to discriminate building regions from light detection and ranging (LiDAR) data using the following steps. First, digital surface model (DSM) data are obtained by the interpolation of a LiDAR point cloud, and a top-hat transform is introduced to remove outliers. Second, a vegetation mask is derived by using the entropy and normalized difference vegetation index extracted from the DSM and aerial imagery, respectively. Third, a stack of nested binary images is generated by slicing the DSM data into different levels, and in each level, the connected components are acquired by using vegetation-mask-based connected analysis. Finally, a tree structure is constructed using a max-tree algorithm, and then building regions are derived by analyzing the area difference of the corresponding nodes of the tree in adjacent levels. The proposed VMCF algorithm is validated using three test areas provided by the German Society for Photogrammetry, Remote Sensing and Geoinformation. The experimental results show that building regions in the LiDAR data can be effectively detected by the proposed method, and the detection rates of three test areas are 89%, 91.8%, and 90.9%, respectively.