PLANNING MULTIPLE OBSERVATIONS FOR OBJECT RECOGNITION

Most computer vision systems perform object recognition on the basis o f the features extracted from a single image of the object. The proble m with this approach is that it implicitly assumes that the available features are sufficient to determine the identity and pose of the obje ct uniquely. If this assumption is not met, then the feature set is in sufficient, and ambiguity results. Consequently, much research in comp uter vision has gone toward finding sets of features that are sufficie nt for specific tasks, with the result that each system has its own as sociated set of features. A single, general feature set would be desir able. However, research in automatic generation of object recognition programs has demonstrated that predetermined, fixed feature sets are o ften incapable of providing enough information to unambiguously determ ine either object identity or pose. One approach to overcoming the ina dequacy of any feature set is to utilize multiple sensor observations obtained from different viewpoints, and combine them with knowledge of the 3-D structure of the object to perform unambiguous object recogni tion. This article presents initial results toward performing object r ecognition by using multiple observations to resolve ambiguities. Star ting from the premise that sensor motions should be planned in advance , the difficulties involved in planning with ambiguous information are discussed. A representation for planning that combines geometric info rmation with viewpoint uncertainty is presented. A sensor planner util izing the representation was implemented, and the results of pose-dete rmination experiments performed with the planner are discussed.