Progressively Extracting Accurate Building Roofs from Airborne LiDAR Data

Extracting building roofs from LiDAR data is a key processing step of 3D building model reconstruction. In this paper, we present a progressive method for accurately extracting building roofs with complex shapes from airborne LiDAR data. Our method first segments original LiDAR data into a set of rough roofs, with larger area and distinct edges, by the region growing algorithm with a normal threshold and a curvature threshold. Accurate roofs are extracted based on estimating plane equations of the rough roofs using the principal component analysis technique. Our method finally employs the Random Sample Consensus algorithm(RANSAC) to extract smaller roofs from the LiDAR data removed the points belong to the extracted roofs. Experimental results show that the method can robustly extract accurate building roofs in a progressive way from the sampled data by adjusting the thresholds dynamically.