Sign Language Recognition Through Video Frame Feature Extraction using Transfer Learning and Neural Networks
NagaJyothi Devabathini, P. Mathivanan · 2023
Effective communication is essential in a world where technology is connecting people more and more, particularly for those who primarily communicate through sign language. In this study, video datasets are used to systematically explore sign language recognition. Through transfer learning, we leverage the power of popular pre-trained models such as ResNet50, ResNet101, VGG16, InceptionV3, and MobileNetV2 to break down video sequences into individual frames, extracting complex features and labeling them to enable accurate language interpretation. To synthesize these features and labels, our project uses a simple neural network, which allows for precise sign language identification. The results showed subtle differences in performance, with InceptionV3 being the most stable model after 20 epochs with an accuracy of 97.82%. By means of a comparative examination of accuracy levels among several pretrained models, this study offers significant perspectives on the best frameworks to promote inclusive communication and advance assistive technologies.