Full text of this paper is not available in the ETD Center. Copies may be available for inter-library loan from University of Cincinnati or may be available for purchase from Proquest/UMI.

Degree

PhD, University of Cincinnati, Engineering : Electrical Engineering, 2000.

Abstract

This dissertation is directed towards the development of an inexpensive and non-intrusive gesture recognition system. Gesture plays a significant role in human communication, and the recognition of gestures is central to developing computers which allow their users to interact in a more natural way. The proposed methodology is based on modeling the hand with a set of identically parameterized regions. These regions are chosen to both minimize the total number of regions as well as the total number of parameters used. For the purposes of this dissertation, the disk region is used, consisting of only two parameters (center and radius) while allowing accurate modeling of the hand with as few regions as possible. To preserve the structural relation between the model components, a tree representation is added to the model. Using this refined model, consisting of both the disks and a connectivity preserving tree structure, the movements comprising a given gesture can be tracked over time. Using the local property of the disk locations and the global property of the interconnectivity, the state of the hand is represented. Edit distances between models found at successive time instants are used to detect the motion of the hand. This signature – the model and series of vectors indicative of the disk movements – is used as the basis for the recognition of the gesture. The recognition process is performed using a feed-forward neural network architecture with the addition of memory neurons. As inexpensive cameras become a standard peripheral on desktop computers, it is anticipated that the methodology proposed herein and extensions of it will allow for more natural human-computer interaction. Areas in which human-computer interaction would be enhanced and occasionally promoted by the addition of gestures are numerous. Such applications include using sign language for computer input and learning, overcoming physical limitations which preclude the use of a keyboard, interacting with virtual environments, and allowing for computer avatars to exhibit the same gestures that humans do.

Keywords

Neural Networks; Human-Computer Interaction; Pattern Recognition

Advisor

Ravi Kothari

Document number: ucin973088031
Permalink: http://rave.ohiolink.edu/etdc/view?acc_num=ucin973088031