Automatic registration of laser point clouds of urban areas
Many tasks in airborne laserscanning require the registration of different scans of the same object. Especially data acquired in urban environments with buildings viewed obliquely from different directions need to be aligned. In this paper we propose a method to filter these point clouds based on different techniques to speed up the computations and achieve better results with the ICP algorithm. A statistical analysis is used and planes are fitted to the data wherever possible. In addition to this, we derive extra features that are used for better evaluation of point-to-point correspondences. These measures are directly used within our extension of the ICP method instead of pure Euclidean distances. Both the intensity of reflected laser pulses and normal vectors of fitted planes are considered. The extended algorithm shows faster convergence and higher stability. We demonstrate and evaluate our approach by registering four data sets that contain different oblique views of the same urban region.