Sistem Klasifikasi Bahasa Isyarat secara Realtime menggunakan SSD MobilNet dengan Tensorflow2

Abstract

Sign language is the primary communication tool for the deaf community. However, limited public understanding creates communication barriers. This study develops a real-time sign language classification system using the SSD MobileNetV2 architecture based on TensorFlow. The dataset consists of 2,000 BISINDO hand gesture images across 10 classes. The system was tested under offline and realtime conditions and at various distances (0cm to 150cm). The highest accuracy reached 90.2% at 50cm and 96% accuracy in realtime for the "Wah, Keren" class. This system significantly contributes to accessible technology for the deaf community.