Realtime Pointing Gesture Recognition and Applications in Multi-user Interaction

摘要

Pointing is a common gesture of human. Indeed, people tend to involve pointing action in their daily activities using not only bare hand but also with gloves or a kind of pointers like pens, rulers, long sticks, batons, etc. Thus, in this article, the authors propose a new concept of interaction that is centered by the natural gesture of human and a method to detect it under various circumstances. Different from some common approaches which rely on predefined skin color or markers, the proposed method can segment and detect any pointer tip and allow multiple objects to be processed at a time. The method can run with the average accuracy of 91.0%. In case of multiple users using different pointing objects, the accuracy is slightly reduced to 87.9%. The running time is at most 17.14 ms for 9 objects being processed in parallel, and thus can be applied for real time constraints.