Terrestrial laser scanning instruments are coming more and more into operation. Up to now vendors of laser scanners mainly use manual or semi-automated registration techniques combined with artificial targets to register single scans. The automatic matching of multiple scans without additional targets is still a research topic in the field of terrestrial laser scanning. Many different proposals and algorithms have already been presented to solve this task. Nevertheless, a strong demand remains for fast and robust algorithms for the registration of multiple scans. The problem is difficult to solve and the process is often divided into two stages. First a coarse matching is done in order to determine a pre-alignment of the scanned surfaces. Then a fine matching algorithm is used to achieve more accurate results. Robust algorithms like the well known iterative closest point (ICP) algorithm and lots of variants already exist for the fine matching, whereas the existing methods for the pre-alignment of the scan data are often rudimental and limited. In this paper a proposal for automatic registration of terrestrial laser scanning data using extended Gaussian images is introduced. The method is placed in the coarse matching stage of the registration process and can be used to determine the rotation component between different scan positions. Therefore normal vectors of local or segmented planes of the input data are used. 1

Terrestrial laser scanning systems have become widely available during the past years. Raw data acquired by such systems typically consists of separate overlapping datasets – each in its own local coordinate system. Applications that need data from more than a single scan position therefore must be preceded by a registration of all scans into a common geometric reference frame. In this paper, a novel method for the automatic and marker-free coarse registration of terrestrial laser scan data is presented. It is based on matching planes in object space and is thus especially suitable for scenarios that are dominated by planar structures such as built-up areas. First, suitable planes are extracted from the raw point cloud in a robust way. Then, the automatic coarse registration is carried out based on correspondences of single plane pairs. Results are shown for test data constisting of 26 datasets of a small village. 1.1 Motivation 1