Multiview registration of 3D scenes by minimizing error between coordinate frames

Abstract

This paper addresses the problem of large-scale multiview registration of range images captured from unknown viewing directions. To reduce the computational burden, we separate the local problem of pairwise registration on neighboring views from the global problem of distribution of accumulated errors. We define the global problem as an optimization over the graph of neighboring views, and we show how the graph can be decomposed into a set of cycles such that the optimal transformation parameters for each cycle can be solved in closed form. We then describe an iterative procedure that can be used to integrate the solutions for the set of cycles across the graph of views. This method for error distribution does not require point correspondences between views, and can be used to integrate any method of pairwise registration or robot odometry.

A review of recent range image registration methods with accuracy evaluation

“…However, classic multi-view algorithms are not the best option because in large surfaces there are lots of views without overlapping area. So, some authors proposed to build adjacent graphs to determine loops which are further register using a multi-view algorithm [44]. Fig.…”

Eliminating conditionally independent sets in factor graphs: A unifying perspective based on smart factors

“…Therefore, we compute {R i,i+1 } from (19) and backsubstitute the optimal value in the expression of the factor. This problem has been explored with a different application in [6] and [27]. Therefore we know that the optimal solution can be computed for fixed (unknown) R u and R v as:…”

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

“…However, as we also observed in our experiments, some criticality remain when handling multiple closure problems (from the acquisition of complex objects), as well as when the number of views to align increases up to the point that the underlying heuristic fails to converge to the global minimum of the error function. Quaternion representation of rigid rototranslation transformations [35] is exploited in several global registration works [36], [37], [38]. Benjemaa and Schmidt [36] demonstrate that the optimal translation can be decoupled and solved independently from the optimal rotation.…”

“…Their approach is based on an iterative method where rotation solutions are found based on the movement of one view at a time while keeping the others fixed. The same decoupling is exploited in the work of Sharp et al [37] where optimization over the graph of neighboring views in a quaternion space is done and closed form solutions are obtained on the cycles of a graph decomposition. The method does not require the computation of point correspondences and can be combined with any pairwise alignment algorithm to generate the estimates of relative motion between each pair of views.…”