Implementation of MFCC based hand gesture recognition on HOAP-2 using Webots platform

Abstract

Hand gestures are the only means of communication and interaction for hearing impaired. This paper proposed a computer vision based technique to identify hand gestures from library of Indian Sign Language (ISL) gestures and Sheffield Kinect Gesture (SKIG) Dataset. Mel Frequency Ceptral Coefficients (MFCC) is used as feature vector due to its high quality of discriminating power in different classes. Minimum distance classifier (Euclidean distance metric) is used for classification of different gestures of a same person in two different lighting conditions, yellow light and white light as well as on SKIG data set. Performance of the proposed technique is evaluated on ten types of ISL gestures (5 are dynamic and 5 are static gestures) and five types of SKIG Kinect gestures and compared with the existing techniques which are also performed on SKIG gesture dataset and ISL dataset. Comparative analysis of our proposed method is performed with the existing method. Performance analysis of our proposed method shows better results than the orientation histogram based technique. Here ISL gestures are simulated on HOAP-2 Robot in Webots platform, for establishing interaction between robot and human.