Features Fitting using Multivariate Gaussian Distribution for Hand Gesture Recognition
Mokhtar Mohammed Hasan · 2012
Gesturing and posturing are two common tools used by the human to support his oral language during communicating with others which helped and proved its ability to deliver the message easily and correctly especially when those were on some sight distance, in this paper we have implemented a novel approach for providing such intuitive interface but this time will be used between human and human-made machine for human computer interaction purposes that helps the hearing impaired people as well, our approach based on feature distribution using multivariate Gaussian distribution function for finding a permanent remedy for translation, scaling, as well as rotation as one pack, we have focussed mainly on the rotation perturbation which solved normally by providing tenfold, hundredfold and even sometimes thousand fold of training gestures in training phase, we have headed a different direction by focusing on lowering this simmering number of training patterns and producing a unified set of features for modelling the hand class, these features are controlled by the direction of hand object that is extracted using our direction analysis algorithm, central moments are used herein for feature vector representation and static gesture recognition that proves its robustness, we have achieved a remarkable recognition rates especially with few number of training samples which is our aim of this study with a recognition time of 0.794 second.