Predicting Object Recognition Performance under Data Uncertainty, Occlusion and Clutter

W e present a novel method f o r predicting the per-formance of a n object recognition approach in the pres-ence of data uncertainty, occlusion and clutter. The recognition approach uses a vote-based decision crite-rion, which selects the object/pose hypothesis that has the m a x i m u m number of consistent features (votes) with the scene data. The prediction method deter-mines a fundamental, optimistic, limit o n achievable performance by any vote-based recognition system. It captures the structural similarity between model ob-jects, which is a fundamental factor in determining the recognition performance. Given a bound o n data uncertainty, we determine the structural similarity be-tween every pair of model objects. This is done by computing the number of consistent features between the two objects as a funct ion of the relative trans-formation between them. Similarity information is then used, along with statistical models f o r data distor-tion, t o estimate the probability of correct recognition ( P C R) as a funct ion of occlusion and clutter rates. The method is validated by comparing predicted PCR plots with ones that are obtained experimentally.