Hand gesture recognition based on dimensionality reduction of histogram of oriented gradients

Abstract

Hand Gesture Recognition (HGR) system has become essential tool for deaf-dumb people to interact with normal users via computer system. This paper proposes robust and fast system for HGR that is based on dimensionality reduction of histogram of oriented gradients feature vectors by applying principal component analysis without losing performance, besides reducing computational cost and memory requirements. Multi-class Support Vector Machine (SVM) and k-Nearest Neighbors (KNN) classifiers are used to classify the hand gestures. The proposed algorithm achieves average recognition rate of 97.69% under different hand poses and complex background with changes in lightning. Our proposed algorithm reduces gesture matching computational cost and memory requirements by 98.6%. Experimental results show that average accuracy with KNN classifier are better than with SVM classifier. The results also show that our descriptor is robust against multiple variations such as rotation, scale, translation, and lighting while provides good performance.