Global registration of large collections of range images with an improved Optimization-on-a-Manifold approach

Abstract

Concurrently obtaining an accurate, robust and fast global registration of multiple 3D scans is still an open issue for modern 3D modeling pipelines, especially when high metric precision as well as easy usage of high-end devices (structured-light or laser scanners) are required. Various solutions have been proposed (either heuristic, iterative and/or closed form solutions) which present some compromise concerning the fulfillment of the above contrasting requirements. Our purpose here, compared to existing reference solutions, is to go a step further in this perspective by presenting a new technique able to provide improved alignment performance, even on large datasets (both in terms of number of views and/or point density) of range images. Relying on the 'Optimization-on-a-Manifold' (OOM) approach, originally proposed by Krishnan et al., we propose a set of methodological and computational upgrades that produce an operative impact on both accuracy, robustness and computational performance compared to the original solution. In particular, always basing on an unconstrained error minimization over the manifold of rotations, instead of relying on a static set of point correspondences, our algorithm updates the optimization iterations with a dynamically modified set of correspondences in a computationally e↵ective way, leading to substantial improvements in terms of registration accuracy and convergence trend. Other proposed improvements are directed to a substantial reduction of the computational load without sacrificing the alignment performance. Stress tests with increasing views misalignment allowed to appreciate the convergence robustness of the proposed solution. Eventually, we demonstrate that Preprint submitted to Image and Vision Computing March 16, 2014 for very large datasets a further computational speedup can be reached by the adoption of an hybrid (local heuristic followed by global optimization) registration approach.

DenseMatch: a dataset for real-time 3D reconstruction

“…Then we fixed by hand the failed alignments that occurred during the process by running again the pipeline with user-selected control points for the critical point clouds. Finally, we refined the result in a global fashion by leveraging on an optimization-on-a-manifold framework [8] . Eventually, the final set of point clouds is properly aligned, thus constituting the ground truth for different kinds of tests on 3D object registration or for training/fine-tuning models to cope with small-scale 3D objects and dense acquisitions.…”

Automatic Multiview Alignment of RGB-D Range Maps of Upper Limb Anatomy

“…In contrast, lightweight configurations can be achieved by integrating a limited number of sensors, although this approach necessitates multiple scanner placements around the patient’s arm to capture the complete anatomy [ 15 ]. This technique is more flexible and adaptable to different patients’ needs but requires accurate registration of the acquired multiple views into a common reference frame, which is conventionally composed of two alignment steps [ 16 , 17 ]: a local coarse alignment and a global fine registration. The first step is based on manually identifying common landmarks on different scans.…”