Real time vision-based hand gesture estimation for a human-computer interface

In this paper we discuss a technique for real-time hand gesture estimation by user interfaces. First, an image processing algorithm for hand gesture estimation is formulated assuming the hand to be a non-rigid body. A method is then proposed for real-time hand gesture estimation using pipelined parallel processing in real-time. With the proposed method, erroneous finger extraction is eliminated using global hand features to ensure stable gesture recognition. The influence exerted by occlusions on hand gesture estimation is examined and it is shown that this influence can be eliminated by an appropriate arrangement of multiple cameras. The efficiency of the proposed method is verified by building a user interface environment incorporating a prototype setup. In so doing, it is made clear that an acceptable interface environment may be implemented even without complete solution of problems related to occlusions or hand orientation. © 1997 Scripta Technica, Inc. Syst Comp Jpn, 28(7): 76–88, 1997