Gesture recognition toolkit using a Kinect sensor

Computational modeling of human behavior has become a very important field of computer vision. Gesture recognition allows people to interact with machines in a natural way without the use of dedicated I/O devices. This paper presents a simple system that can recognize dynamic and static gestures using the depth map and the higher level output (skeleton and facial features) provided by a Kinect sensor. Two approaches are chosen for the recognition task: the Dynamic Time Warping Algorithm is used to recognize dynamic gestures, while a Bayesian classifier is used for the static gestures/postures. In contrast with some specialized methods presented in the literature, the current approach is very generic and can be used with minimal modification for recognizing a large variety of gestures. As a result, it can be deployed in a multitude of fields from security (monitoring rooms and sending alarm signals), medicine (helping people with physical disabilities) to education and so on. The tests results show that the system is accurate, easy to use and highly customizable.