Accommodating uncertainty in pixel-based verification of 3-D object hypotheses

Abstract We present a novel technique for verifying 3-D object hypotheses using an intensity image. Verification is performed through pixel-wise comparison of edge images corresponding to scene data and hypothesized model object. We accommodate the uncertainties involved in this process, which correspond to bounded positional errors of scene and model edge pixels, by dilating the scene edge image. An analytical framework is presented for formally determining the extent of dilation under the perspective projection. Performance of the technique is demonstrated using real and synthetic data.