Spectral 6DOF Registration of Noisy 3D Range Data with Partial Overlap

Abstract

We present Spectral Registration with Multilayer Resampling (SRMR) as a 6 Degrees Of Freedom (DOF) registration method for noisy 3D data with partial overlap. The algorithm is based on decoupling 3D rotation from 3D translation by a corresponding resampling process of the spectral magnitude of a 3D Fast Fourier Transform (FFT) calculation on discretized 3D range data. The registration of all 6DOF is then subsequently carried out with spectral registrations using Phase Only Matched Filtering (POMF). There are two main aspects for the fast and robust registration of Euler angles from spherical information in SRMR. First of all, there is the permanent use of phase matching. Second, based on the FFT on a discrete Cartesian grid, not only one spherical layer but also a complete stack of layers are processed in one step. Experiments are presented with challenging datasets with respect to interference and overlap. The results include the fast and robust registration of artificially transformed data for ground-truth comparison, scans from the Stanford Bunny dataset, high end 3D laser range finder (LRF) scans of a city center, and range data from a low-cost actuated LRF in a disaster response scenario.

On evaluating consensus in RANSAC surface registration

“…Another group of methods attempt to extend the notion of Phase Correlation [DCM87], which has been successfully used for image alignment, to 3D point clouds. Usually, the input point clouds are resampled onto a regular grid, which is then transformed into frequency domain, where the aligning rotation can be found [BB13]. Unfortunately, the extension only works in a certain range of angles, and thus it cannot be applied for a general global registration.…”

A hough transform based scan registration strategy for Mobile Robotic Mapping

“…To improve the stability of HSM3D, a set of candidate solutions are produced, hence the HSM3D is computationally expensive and an extra evaluation step is necessary to pick out the correct solution. A more recent frequency-based approach SRMR [17] resamples the spectral magnitude of 3D Fast Fourier Transform (FFT) calculated on discrete Cartesian grids of the 3D range data to detach the 3D rotation from 3D translation. This method works only within a restricted range of roll and pitch changes between scans.…”

“…That is because our method extracts not only the points on the ground but also the points at the bottom of the objects, and this case gives full play to the RANSAC method. At worst, we could assume that the normal vector of ground surface parallels the z axis and regard the roll and pitch as an undesirable noise, like the algorithm in [17] dose, and then compensate the roll and pitch on the strength of Fig. 6.…”

Automatic registration of optical aerial imagery to a LiDAR point cloud for generation of city models

“…However, the algorithm requires an initial registration in order to converge. Existing methods for initial registration can grouped into feature-based (Li and Guskov, 2005;Gelfand et al, 2005;Rusu et al, 2009;Mérigot et al, 2009) and featureless methods (Dold, 2005;Pathak et al, 2010;Lucchese et al, 2002;Bulow and Birk, 2013).…”