Hypothesis Verification in Model-Based Object Recognition with a Gaussian Error Method

The use of hypothesis verification is recurrent in the model based recognition literature. Small sets of features forming salient groups are paired with model features. Poses can be hypothesised from this small set of feature-to-feature correspondences. The verification of the pose consists in measuring how much model features transformed by the computed pose coincide with image features. When data involved in the initial pairing are noisy the pose is inaccurate and the verification is a difficult problem. In this paper we propose a robust hypothesis verification algorithm, assuming data error is Gaussian. Previous approaches using gaussian error model start from an initial set of correspondences and try to extend it feature by feature. This solution is not optimal. In this paper, the opposite strategy is adopted. Assuming the right pose belongs to a known volume of the pose space (including the initial pose) we take into account all of the correspondences compatible with this volume and refine iteratively this set of correspondences. This is our main contribution. We present experimental results obtained with 2D recognition proving that the proposed algorithm is fast and robust.