A Novel Approach to View Planning in Shape Acquisition from Range Data

Abstract

We present an image-based approach to the determination of the next best view (NBV) in the incremental construction of object models from multiple range images. No \textit{a priori} constraints are imposed on object shape, which is considered completely unknown. Rather than reconstruct a 3-D model like most current methods do, we use image-based representations to model the captured range information. Our method requires only conventional graphics hardware, and since it uses image-based rendering, which is independent of object complexity, it makes near-real-time NBV determination for arbitrarily complex objects possible in practice.