Using Polar Grid for Building Extraction in Terrestrial Laser Scanning Data

Abstract

In this paper, we propose a method for building extraction, mainly focusing on the large density variety in terrestrial laser scanning (TLS) data. The density of projected points is calculated based on the polar grid instead of the commonly used regular grid and adaptive thresholds are generated for each cell to filter non-building cells and avoid parameter tuning. The points in the remaining cells are grouped into different clusters and the planar ratio of each cluster is calculated based on dimensionality features. Finally, the points in the clusters with planar ratio larger than 75% are recognized as building points. The method is compared with a regular grid based method in an urban TLS data with the buildings distributed at a distance of 60 to 350m and the results show that the proposed method has a better performance on the detection of buildings which are more than 200 meters away in this data.