BUILDING BOUNDARY EXTRACTION FROM LIDAR DATA USING A LOCAL ESTIMATED PARAMETER FOR ALPHA SHAPE ALGORITHM

Abstract

The α-shape algorithm is a very common option to extract building boundaries from LiDAR data. This algorithm is normally executed in 2D space considering a parameter α as a binary classifier which controls the distinctiveness of points whether or not they belong to the object boundary. For point cloud data, this parameter is directly related to the local point density and the level of detail of building boundaries. Studies that have explored this concept usually consider a unique parameter α to extract all buildings in the dataset. However, the point density can have a considerable variation along the point cloud and, in this case, the use a global parameter may not be the best choice. Alternatively, this paper proposes a data-driven method that estimates a local parameter for each building. The method evaluation considered six test areas with different levels of complexity, selected from a LiDAR dataset acquired over the city of Presidente Prudente/Brazil. From the qualitative and quantitative analysis, it could be seen that the proposed method generated better results than when a global parameter is used. The proposed method was also able to withstand density variation among the LiDAR data, having a positional accuracy around 0.22 m, against 0.40 m of global parameter.