Real-time dynamic visual gesture recognition in human-robot interaction

This paper tackles a common problem in human-robot interaction: recognizing the intentions of a human in an intuitive way. We present a system that is able to recognize dynamic human gestures in an interaction scenario where a humanoid robot visually observes the behavior of a human. This allows for a natural human-robot communication where no markers or technical devices are necessary on the side of the human interactor. The system not only recognizes previously learned gestures, but is also able to categorize and learn new gestures in an unsupervised manner. The proposed approach stands out due to its low computational cost and therefore can be used with the potentially slow embedded hardware of a humanoid robot. To demonstrate the possibilities of the approach we arranged a human - humanoid interaction game which consists of an alternating gesture-based communication.