Recognition of global hand gestures using self co-articulation information and classifier fusion

Abstract

Hand gestures are a natural way of communication in our everyday life. Gestures provide a user friendly alternative of various types of devices available in our surroundings such as keyboard, mouse, and joysticks. This paper presents a hand gesture recognition system based on new set of features along with the classifier combining techniques to enhance the performance of the existing system. The study reveals that the unwanted movement within a gesture (self co-articulation) may be used as one of the complementary feature to improve the performance of the system. The self co-articulation was detected from the gesture trajectory by exploiting velocity information along with the pause information. The proposed new set of features (such as position of the hand, ratio feature and first half trajectory features) along with the self co-articulated features were used to define the feature space. Individual models as well as hybrid models were developed using the classifiers such as artificial neural network (ANN), multiclass support vector machine (SVM), k-nearest neighbours (k-NN) and Naive Bayes model. The models based on conditional random field (CRF) and hidden conditional random field (HCRF) were used to develop the baseline system for the present study. The experimental results suggest that the proposed new set of features combined with the hybrid classifier provides an improvement in the accuracy of the system by 14.58 and 8.41 % as compared to CRF and HCRF based models respectively. An analysis of variance test has also been conducted to validate the statistical significance of results.