| Summary: | Thesis (M.Eng.)--Memorial University of Newfoundland, 1998. Engineering and Applied Science Bibliography: leaves 95-97. The development of a reliable multi-modal human-machine interface has many potential applications. The interface with a personal computer has become very common yet many disabled users have limited access due to the restrictiveness of the current interface. An improved interface would improve the quality of life for disabled users and has applications in controlling machinery in an industrial setting. Many different types of gestures ranging from head gestures, headpointing, hand and arm gestures are being investigated. A wide variety of classification techniques are available. These techniques range from simple clustering routines to complex adaptive routines. This work compares the recognition results of four pattern recognition techniques, the k-nearest neighbor, a Mahalanobis distance classifier, a rule based classifier and hidden Markov models. The techniques were tested on a set of six hand gestures captured using The Flock of Birds data collection system. The best average recognition result was 97% obtained from the k-nearest neighbor classifier, the Mahalanobis distance classifier had an average recognition rate at 92%, the rule based classifier had an average recognition rate at 89% and the hidden Markov models had the lowest average recognition results at 83%. The hidden Markov models are the most complex of the four techniques studied. Although the average recognition results were lower, they are rich in mathematical structure and can be used to model very complex observations.
|
|---|