Registering multiview range data to create 3D computer objects

Abstract

This research deals with the problem of range image registration for the purpose of building surface models of three-dimensional objects. The registration task involves nding the translation and rotation parameters which properly align overlapping views of the object so as to reconstruct from these partial surfaces, an integrated surface representation of the object. The approach taken is to express the registration task as an optimization problem. We de ne a function which measures the quality of the alignment between the partial surfaces contained in two range images as produced by a set of motion parameters. This function computes a sum of Euclidean distances between a set of control points on one of the surfaces to corresponding points on the other. The strength of this approach resides in the method used to determine point correspondences across range images. It is based on reversing the range nder calibration process, resulting in a set of equations which can be used to directly compute the location of a point in a range image corresponding to an arbitrary point in three-dimensional space. A stochastic optimization technique, Very Fast Simulated Reannealing (VFSR), is used to minimize the cost function. Dual-view registration experiments yielded excellent results in very reasonable computational time. A multiview registration experiment was also performed, but a large processing time was required. A complete surface model of a typical 3D object was then constructed from the integration of its multiple partial views. The e ectiveness with which registration of range images can be accomplished makes this method attractive for many practical applications where surface models of 3D objects must be constructed.

Real-Time RGB-D Mapping and 3-D Modeling on the GPU Using the Random Ball Cover

“…Compared to related methods based on projective data association [5] that primarily consider the surface geometry for finding corresponding points, our approach allows to incorporate multiple complementary sources of information (in our case geometry and photometry) into the nearest neighbor search. Furthermore, explicitly performing a nearest neighbor search according to a point signature potentially allows to extend the framework to handle large misalignments by a feature-based initial pre-alignment [3].…”

“…In particular, the KinectFusion framework [23,31] has gained popularity in the field of 3-D reconstruction. The fundamental core of this framework is based on the work of Rusinkiewicz et al [35], combining projective data association [5] and a point-to-plane metric [9] for rigid ICP surface registration and sensor pose estimation, respectively. While the original work was limited to a frame-to-frame alignment [35], KinectFusion tracks the depth frame against a globally fused implicit surface model of the observed scene [10].…”

Study on optical measurement technology for automotive covering parts

“…ICP is a commonly used method for point cloud registration [4]. But the traditional ICP method is easily affected by the initial position, easy to fall into local optimum.…”

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

“…Some works are based on out-of-core computed correspondence sets, others foresee to update the correspondences in an iterative manner, others try to avoid using them at all. Among representative early works on global registration, Blais and Levine [25] minimize an Euclidean distance cost function, calculated on sets of control points, by simulated annealing. Later, Silva et al [26] adopt a similar approach exploiting genetic algorithms with a surface interpenetration measure.…”