Multilingual sign language translation using image-based classification models for Bangla, American, and Indian languages with ResNet and EfficientNet

Abstract

This study proposes a multilingual sign language recognition system for Bangla Sign Language (BdSL), American Sign Language (ASL), and Indian Sign Language (ISL). It uses ResNet50 and EfficientNetB0 to recognize static hand gestures from images. ResNet50 performs well in identifying visually similar signs, while EfficientNetB0 offers lower computational cost, making it suitable for portable applications. The study highlights limitations in recognizing continuous gestures and recommends future research using video data, LSTM or transformer-based models, and real-time mobile/web applications.