Hand Gesture Detection using Transfer Learning with Deep Neural Networks

Large segments of the population value sign language as one of the many languages used for communication. Each sign in each sign language differs in terms of hand shape, motion profile, and positioning of the hand. These differences will have a substantial influence on the accuracy and reliability of the identification process. Therefore, one of the key techniques for detecting gestures is sign language identification utilizing deep learning. In this paper, we have considered four major algorithms of Image classification EfficientNet- B0, GoogleNet, ResNext 50 32x4d and MobileNet-V2 which were tested on GPU system. To determine the optimum model that uses the fewest GPU cores, many factors have been considered, including training time, training accuracy, total loss and inference. GoogleNet performance is better as compared to other models.