Marker-free Registration of Terrestrial Laser Scans Using the Normal Distribution Transform

The registration of scan data often uses special markers which are placed in the scene. This leads to a reliable registration but the method is not very efficient. Therefore, we search for a registration method which works without markers. There are methods like the iterative closest point (ICP) algorithm which calculate the registration on the basis of the data itself. However, these algorithms have a small convergence radius and therefore a manual pre-alignment is necessary. In this paper, we explore a registration method called the normal distribution transform (NDT) which does not require markers, has a larger convergence radius than ICP and a medium alignment accuracy. The NDT was initially proposed in robotics for single-plane horizontal scans. We investigate three modifications to the original algorithm: a coarse-tofine strategy, multiple slices, and iterative solution using the method of Levenberg-Marquardt. We apply the modified algorithm to real terrestrial laser scanner data and discuss the results.