Date of Award
Master of Science
Vasant G. Honavar
The increasing number of new and complex computer-based applications has generated a need for a more natural interface between human users and computer-based applications. This problem can be solved by using hand gestures, one of the most natural means of communication between human beings. The difficulty in deploying a computer vision-based gesture application in a non-controlled environment can be solved by using new hardware which can capture 3D information. However, researchers and others still need complete solutions to perform reliable gesture recognition in such an environment.
This paper presents a complete solution for the one-hand 3D gesture recognition problem, implements a solution, and proves its reliability. The solution is complete because it focuses both on the 3D gesture recognition and on understanding the scene being presented (so the user does not need to inform the system that he or she is about to initiate a new gesture). The selected approach models the gestures as a sequence of hand poses. This reduces the problem to one of recognizing the series of hand poses and building the gestures from this information. Additionally, the need to perform the gesture recognition in real time resulted in using a simple feature set that makes the required processing as streamlined as possible.
Finally, the hand gesture recognition system proposed here was successfully implemented in two applications, one developed by a completely independent team and one developed as part of this research. The latter effort resulted in a device driver that adds 3D gestures to an open-source, platform-independent multi-touch framework called Sparsh-UI
Bonansea, Lucas, "3D Hand gesture recognition using a ZCam and an SVM-SMO classifier" (2009). Graduate Theses and Dissertations. 10829.