Deep Learning in Autoencoder Framework and Shape Prior for Hand Gesture Recognition

Abstract

AbstractGesture is a form of nonverbal communication and is widely used in everyday life. Gesture recognition system facilitates human computer interaction thus creating an interface, which helps computer to understand human body language. In this article, an automated hand gesture recognition system is proposed using deep learning framework. The proposed scheme classifies the input hand gestures, each represented by a set of feature vector into some predefined number of gesture classes. The deep learning algorithm is used here to recognize hand gestures. The proposed scheme consists of mainly three stages: preprocessing, feature extraction, and classification. The images of hand gestures are stored in a database and are preprocessed to get into suitable form for further processing. The histogram-oriented gradient (HOG) feature vectors are then extracted from the preprocessed gesture images. The feature vectors are trained using deep learning techniques based on which the hand gesture from test set is classified into different classes. The proposed scheme is tested on three different hand gesture databases. The results obtained by the proposed scheme are compared with different state-of-the-art techniques and are found to be better as compared to the four existing state-of-the-art techniques. The performance of the proposed scheme is validated by using different percentage of the training samples with k-fold cross validation.KeywordsHuman-Machine Interface (HMI)Histogram of oriented gradients (HOG)Deep neural network (DNN)