A Deep Convolutional Neural Network Approach for Static Hand Gesture Recognition

Abstract The communication barrier and the hearing majority are the key social concerns of the deaf-dumb community that prevent them from accessing the basic and essential services of the life. Eventhough the problem has been addressed with the innovations in automatic sign language recognition, an adequate solution has not yet been attained due to a number of challenging factors. Most of the existing works try to develop vision based recognizers through classical pattern analysis approach by deriving complex hand crafted feature descriptors from the captured images of the gestures. But the efficiency of those methods are very limited to work with large sign vocabulary captured in complex and uncontrolled background conditions. This paper proposes a methodology for the recognition of hand gestures, which is the prime component in sign language vocabulary, based on an efficient deep convolutional neural network (CNN) architecture. The method has been tested on two publicly available datasets (NUS hand posture dataset and American fingerspelling A dataset) and achieved better recognition accuracies.