Developing task-specific RBF hand gesture recognition

In this paper we develop hand gesture learning and recognition techniques to be used in advanced vision applications, such as the ActIPret system for understanding the activities of expert operators for education and training. Radial Basis Function (RBF) networks have been developed for reactive vision tasks and work well, exhibiting fast learning and classification. Specific extensions of our existing work to allow more general 3-D activity analysis reported here are: 1) action-based representation in a hand frame-of-reference by pre-processing of the trajectory data; 2) adaptation of the time-delay RBF network scheme to use this relative velocity information from the 3-D trajectory information in gesture recognition; and 3) development of multi-task support in the classifications by exploiting prototype similarities extracted from different combinations of direction (target tower) and height (target pod) for the hand trajectory.