Unsupervised gesture recognition system for learning manipulative actions in virtual basketball

For natural human-agent interaction, the recognition of important gestures to facilitate communication is necessary. Recent advances in technology have made it possible to use the human body as a means for this interaction. In this paper, we combine techniques to create a process flow which can discover and extract a body gesture, create a low-dimensional model, and then use it to recognize body gestures provided in real-time. We show that this technique can be applied to a virtual basketball game, an environment which is highly dynamic and where the variation of gestures is high but the number of training samples is low. We demonstrate this technique to the extraction and recognition of a passing gesture.