City Research Online

Abstract

Recent linguistic research has shown that gesturing is an important channel of non-verbal communications that augments meaning to the spoken word. This thesis demonstrates that a hand gesture can be modelled in 2DT (two dimensions and time) as an aperiodic waveform. Fourier analysis of the waveform generates positive and negative sequence harmonic components from which characteristics of the gestures can be recognised. Variability in gesture lengths are confronted by re-sampling the data to a fixed length, from which harmonic components can be effectively compared. Manipulation of the harmonic data gives the gesture data scale and translation invariant properties. Gesture characterisation is revealed by harmonic ‘orientation’ angles and by each harmonic having a unique ‘elliptical-corkscrew’. The first three harmonics are generally sufficient to characterise a gesture. Gesture recognition is accomplished by using clustering techniques on the low order harmonic data to select target gestures for a Probabilistic Neural Network (PNN). The PNN requires minimal training and in association with clustering techniques, can select target gestures to reveal inter-class and intra-class differences of gestures ensembles. The application of Fourier analysis to gesture stimuli shows their predominantly oscillatory and idiosyncratic nature. A reliable technique for recording hand coordinate data was developed that fused skin-colour and motion cues. As a result, objects, which when rank ordered by area, invariably related the most significant object to the dominant gesturing hand. An object selection algorithm corrected for most tracking mistakes. The technique has been successfully extended to track two hands simultaneously. The gesturing of one person was followed when there were three people in the scene. Additional observations of the individual and personalised nature of hand gesture to gesture stimuli by the gesturer, has revealed the potential of a prime alternative method of a vision-based biometric.

Publication Type:	Thesis (Doctoral)
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Departments:	School of Science & Technology > Computer Science School of Science & Technology > School of Science & Technology Doctoral Theses Doctoral Theses

Export

Downloads