Gesture recognition of RGB-D and RGB static images using ensemble-based CNN architecture

The relationship between humans and computers has always been an exciting environment in this thriving period. Gesture-based recognition systems have always been a fascinating and distinct subject with the exponential growth in Computer Vision. It is a very complicated and daunting process to understand human expressions in the form of sign language. Different traditional approaches have increasingly been used to understand sign language, but attain high precision is still a difficult challenge and vision-based finger-spelling identification remains difficult, because of inter-class similarities and intra-class variability. The model considers two modalities, RGB and depth. Finger occlusion and hand shapes accurately detected and can be handled by depth information. A fine-tuned dual-path network is suggested compared to current strategies that process RGB-D images separately, that understands finger-spelling depiction in separate RGB and depth paths and gradually fuses the features acquired from both tracks.