Building Reconstruction from Lidar Data Using Iterative Regularization Approach

This paper presents a scheme for the reconstruction of building models from LIDAR data by using an iterative regularization approach. The proposed scheme comprises three major parts: (1) elevation slicing, (2) boundary regularization, and (3) roof determination. The idea of elevation slicing is similar to the elevation contour map, where each contour line indicates a height level. We select a height interval and extract the building masks in different height levels, where each layer represents the building boundary with equal height. Then, the initial building boundaries are obtained by applying an edge detector. In the boundary regularization, we assume that the building boundaries have two dominant directions. We iteratively apply parallel and orthogonal constraints in building boundary regularization. In the roof determination, the line segments of each building are traced to form a polygon. Then, we shape the roof of each polygon from LIDAR point clouds. A TIN-based region growing is applied to extract the roof planes. The proposed method has been tested with LIDAR data of Hsin-Chu Science-based Industrial Park in northern Taiwan. Experimental results indicate that the proposed scheme reaches high reliability.