Robust segmentation of building points from airborne LiDAR data and imagery

A new approach to segment building points from raw LiDAR data by the integration of LiDAR data and imagery is proposed. The segmentation of coarse building point from LiDAR data is executed by using the algorithm proposed by Zhang et al. [1]. A new dynamic triangle-based algorithm is presented to obtain the precise building points by the refinement of the coarse building data. This new dynamic triangle-based algorithm consists of three steps. Firstly, an erosion operation is applied to the coarse building data, which derives the reliable building points as seed points for the following processing. Secondly, an initial Delaunay triangle-network is constructed based on these seed points and baselines for subsequent triangle-network growing are then selected. Thirdly, suspect points, which are near to building boundaries but not involved in the initial triangle-network, are identified as building points or non-building points by a triangle-network growing strategy considering spatial and spectral information, such as height, slope, roughness, spectrum, and entropy. The experiment demonstrates the proposed approach can derive higher accurate results.