Extraction of Building Roof Boundaries From LiDAR Data Using an Adaptive Alpha-Shape Algorithm

Abstract

The alpha-shape algorithm was developed to extract object shapes in 2-D space; however, the accuracy of the result depends on an appropriate choice of the parameter $\alpha $ . This parameter is directly related to point density and the level of detail of the boundary. Similar approaches usually consider a unique parameter $\alpha $ to extract all buildings in the data set. However, as the point density can vary along the cloud and also along the building, using a global parameter may not be suitable in some situations. This letter proposes an adaptive method to overcome this limitation. It estimates a local parameter $\alpha $ for each edge based on local point spacing. The experiments were performed considering buildings with different levels of complexity, which were selected from two different LiDAR data sets and three densities. Qualitative and quantitative analysis enabled verification of the proposed method, showing good results in cases where significant density variation occurs along the building, and in the extraction of complex buildings such as those composed of convex and concave segments and/or the presence of inner boundaries. The proposed adaptive solution can overcome most limitations of simpler approaches, such as the use of a global parameter or only one parameter per building.