SENSOR MODELING, PROBABILISTIC HYPOTHESIS GENERATION, AND ROBUST LOCALIZATION FOR OBJECT RECOGNITION

In an effort to make object recognition efficient and accurate enough for real applications, we have developed three probabilistic technique s-sensor modeling, probabilistic hypothesis generation, and robust loc alization-which form the basis of a promising paradigm for object reco gnition. Our techniques effectively exploit prior knowledge to reduce the number of hypotheses that must be tested during recognition. Our r ecognition approach utilizes statistical constraints on the matches be tween image and model features. These statistical constraints are comp uted using a model of the entire sensing process-resulting in more rea listic and tighter constraints on matches. The candidate hypotheses ar e pruned by probabilistic constraint satisfaction to select likely mat ches based on the image evidence and prior statistical constraints. Th e resulting hypotheses are ordered most-likely first for verification, thus minimizing unnecessary verifications. The reliability of the ver ification decision is significantly increased by the use of a robust l ocalization algorithm. Our localization algorithm reliably locates obj ects despite partial occlusion and significant errors in initial locat ion estimates. We have implemented these techniques in a system that r ecognizes polyhedral objects in range images. Our results demonstrate accurate recognition while greatly limiting the number of verification s.