Vision Based Hand Gesture Recognition for Human Computer Interaction

Abstract

The hand gesture recognition system is widely used in the development of human-machine interaction. The vision-based hand gesture recognition is achieved by the following steps: preprocessing, feature extraction and classification. The aim of preprocessing stage is to localize the hand region from the image frame. The Laplacian of Gaussian filtering technique along with zero crossing detector is applied on hand gesture images to detect the edges of hand region. This work proposes a novel feature extraction technique, which is based on local histogram feature descriptor (LHFD). The proposed feature is extracted by finding the local histogram of the grayscale gesture image. This technique uses the whole region of the hand to extract the features. The proposed method is invariant to the scaling and illumination. Two standard datasets viz. Massey University gesture dataset (MUGD) and Jochen Triesch static hand posture database are used to evaluate the recognition performance of the proposed technique. The gesture recognition performance of the proposed technique is 99.5%and 95%on Massey University gesture dataset and Triesch dataset respectively, using support vector machine (SVM) classifier. Fourier descriptor along with Convolutional neural network is proposed for continuous frame based hand gesture recognition. Fourier descriptor is used to construct the smooth boundary of each frame of video. CNN is used for classification, the recognition accuracy of the proposed method for dynamic hand gesture recognition is of77:5%