Towards a natural gesture interface: LDA-based gesture separability

AbstractThe goal of this paper is to analyse a method of validating a subset of gesturesto be used as elements of a HCI interface. We investigate the applicability ofLDA for gesture data dimensionality reduction. An Gesture mutual separabilityanalysis of a diverse dataset of 22 natural gestures captured with two motion-capture devices is provided. Fisher criterion is used to produce measures of classseparability and class overlap.Keywords: LDA, gestures, separability, HCI, motion-capture1. IntroductionWith a widespread use of motion-tracking devices, both traditional e.g. com-puter mouse and new Nintendo Wii Remote TM , cell phone accelerometer ar-rays, the importance of motion-based interfaces in Human-Computer Interac-tion (HCI) systems became unquestionable. The commercial success of simplemotion-capture devices led to the development of more robust and versatileacquisition systems, both mechanical e.g. Cyberglove Systems Cyberglove TM ,Measurand ShapeWrap TM , DGTech DG5VHand and optical e.g. MicrosoftKinect