MODEL-BASED 3D OBJECT RECOGNITION USING BAYESIAN INDEXING

This research features the rapid recognition of three-dimensional obje cts, focusing on efficient indexing, A major concern in practical visi on systems is how to retrieve the best matched models without explorin g all possible object matches, We have employed a Bayesian framework t o achieve efficient indexing of model objects, A decision-theoretic me asure of the discriminatory power of a feature for a model object is d efined in terms of posterior probability, Domain-specific knowledge co mpiled off-line from CAD model data is used in order to estimate poste rior probabilities that define the discriminatory power of features fo r model objects, In order to speed up the indexing or selection of cor rect objects, we generate and verify the object hypotheses for feature s detected in a scene in the order of the discriminatory power of thes e features for model objects. Based on the principles described above, we have implemented a working prototype vision system using a feature structure called an LSG (local surface group) for generating object h ypotheses, Our object recognition system can employ a wide class of fe atures for generation of object hypotheses. In order to verify an obje ct hypothesis, we estimate the view of the hypothesized model object a nd render the model object for the computed view, The object hypothesi s is then verified by finding additional features in the scene that ma tch those present in the rendered image, Experimental results on synth etic and real range images show the effectiveness of the indexing sche me. (C) 1998 Academic Press.