Automatic Registration Method of Multi-Source Point Clouds Based on Building Facades Matching in Urban Scenes

Both UAV photogrammetry and lidar have become common in deriving three-dimensional models of urban scenes, and each has its own advantages and disadvantages. However, the fusion of these multi-source data is still challenging, in which registration is one of the most important stages. In this paper, we propose a method of coarse point cloud registration which consists of two steps. The first step is to extract urban building facades in both an oblique photogrammetric point cloud and a lidar point cloud. The second step is to align the two point clouds using the extracted building facades. Object Vicinity Distribution Feature (Dijkman and Van Den Heuvel 2002) is introduced to describe the distribution of building facades and register the two heterologous point clouds. This method provides a good initial state for later refined registration process and is translation, rotation, and scale invariant. Experiment results show that the accuracy of this proposed automatic registration method is equivalent to the accuracy of manual registration with control points.