3D Skeletal Joints-Based Hand Gesture Spotting and Classification

This paper presents a novel approach to continuous dynamic hand gesture recognition. Our contains two main modules: spotting and classification. Firstly, the module pre-segments video sequence with gestures into isolated gestures. Secondly, classification identifies segmented In module, motion of palm fingers are fed Bidirectional Long Short-Term Memory (Bi-LSTM) network for spotting. three residual 3D Convolution Neural Networks based on ResNet architectures (3D_ResNet) one (LSTM) combined efficiently utilize multiple data channels such as RGB, Optical Flow, Depth, positions key joints. The promising performance our is obtained through experiments conducted public datasets—Chalearn LAP ConGD dataset, 20BN-Jester, NVIDIA Dynamic Hand Dataset. outperforms state-of-the-art methods Chalearn dataset.