Applying Similarity Metrics to 3D Acquisition in Structured-Light Systems

Structured light systems use projected light to augment a scene with extra information. The goal of such systems is often the recovery of depth information based on 2D image(s) from one or more viewpoints. Traditionally, the best performing systems use the injected information to "label" points within a scene. These labelings provide the correspondence information needed for 3D reconstruction. In this paper we attempt to demonstrate the feasibility of an alternative, similarity metric based, technique by presenting a simple proof-of-concept system. The technique consists of performing a search for corresponding points from two image sets via a similarity metric and then using the best correspondences as an approximation of the actual point correspondences. Remarkably, we find that this conceptually simple approach can generate data of comparable quality to more complicated techniques. Perhaps the most striking advantage to this approach is that it eliminates the need for camera calibration. We believe these initials results indicate great promise for the application of "data-mining" techniques to the 3D acquisition domain.