APPLICATION OF HOPFIELD NEURAL NETWORKS AND CANONICAL PERSPECTIVES TORECOGNIZE AND LOCATE PARTIALLY OCCLUDED 3-D OBJECTS

The task of recognizing and locating the partially occluded three-dime nsional (3-D) rigid objects of a given scene is considered. The surfac es of 3-D objects may be planar or curved. The 3-D surface information s are captured through range data (depth) map. For recognition we use the principal curvatures, mean curvature and Gaussian curvature as the local descriptions of the surfaces. These curvatures do not change si gnificantly under rotation and translation. Hence they are used as loc al invariant (within certain threshold) features of the surfaces. A ne uro-vision scheme, based upon the matching between the local features of the 3-D objects in a scene and those of the object models is propos ed. Object models are generated using canonical perspectives. The feat ure matching scheme is realized, at two stages, through Hopfield neura l networks. At the first stage edge-points of the scene are matched wi th those of the object models using a Hopfield net. At the second stag e non-edge-points of the scene are matched with those of the object mo dels using another Hopfield net. Compared with conventional object mat ching schemes, the proposed technique provides a more general and comp act formulation of the problem and a solution more suitable for parall el implementation. Finally, the hypothesis generation and verification scheme is proposed for best possible recognition.