Skeleton Based Dynamic Hand Gesture Recognition using LSTM and CNN

Dynamic Hand Gestures offer a natural, non-verbal way of communication that can substitute other communication modalities like verbal speech and script writing. Not only for the voice command, hand gestures also play significant role in Augmented Reality (AR), Virtual Reality (VR) and games. There are some factors like computational cost, flexibility and recognition accuracy that can impact the incorporation of hand gestures in these fields. In this paper, a Dynamic Hand Gesture Recognition (DHGR) approach is propose that is based on Convolutional Neural Network (CNN) and long-short term memory (LSTM). This system is trained to execute the sequence of 3D input data along with the velocities and positions information learned from Leap Motion Controller (LMC).When evaluated and compared with state-of-art DHGR methods, this architecture shows relative high accuracy of 98%.