A Time-Distributed Convolutional Long Short-Term Memory for Hand Gesture Recognition

Abstract

The applications of human-robot interaction have recently raised many research interests, and hand gesture recognition to recognize human gestures in video-based problems is one of them. In the recent decade, deep learning techniques have proven their promising performance in the fields of pattern recognition and computer vision. This study presents an improved version of the Convolutional Neural Network in combination with Long Short-Term Memory for hand gesture recognition. The proposed structure is fully considered in a time-distributed framework to effectively train the network of the frame-level classification. Hence, employing a time-distributed framework, a TD-CNN-LSTM method is developed. Finally, the efficacy of our proposed architecture is evaluated on the recent publicly available GRIT corpus dataset, and we also show that our method outperforms the recent state of the art CNN-LSTM method.