Improvements of RGB-D hand posture recognition using an user-guide scheme

This paper argues that an user-guide plays an important role to make a robust and real-time hand posture recognition system. Instead of designing a new hand posture recognition algorithm, we propose an user-guide scheme which handles issues of environmental conditions as well as appearance-based features for hand detections. This guide estimates heuristic parameters whose values strongly affect the recognition results. The experimental results confirm that even by utilizing a simple hand posture recognition algorithm, the proposed method significantly improves the recognition rate. Without training end-user, the recognition rate achieves only 63%, whereas it obtains 87% with the proposed guide scheme. To guide an end-user, the proposed scheme requires averagely 15 seconds in advance. Therefore, the proposed method is feasible to deploy practical applications, particularly, to control devices in a smart-home such as televisions, game consoles, or lighting systems.