Man-machine interaction: Hand gesture recognition and mimicking with a robotic hand

Robotic vision is a very powerful tool that has to be further explored and refined in order to highly improve robots' behaviour; in particular, it plays a key role in the achievement of a reliable and strong interaction between man and machine. The more natural this interaction becomes for humans, the more popular robotics will grow: being able to communicate through natural speech or gestures, the robot would become a real companion (for both working and recreational purposes) also for non-expert people. In this paper, a preliminary vision-based system for hand gesture recognition and mimicking is presented. The system relies on techniques such as background subtraction, skin detection, SURF feature extraction and matching. Such a system gave positive results both in terms of computational load and execution time, as well as in terms of hit rate of correctly identified human hand gestures. In the paper, each algorithm step is detailed and some experiments are described; moreover discussion about system performance and possible improvements are provided.