Paper Title
Light Weight Network Architecture for Sign Language Recognition
Abstract
Developing sign language recognitionsystems using deep learning models has become a prominent trend.
However, the meaning conveyed by sign language gestures varies across different countries, making it difficult to share open
sign language datasets. Additionally, the complex architecture of deep learning-based sign language recognition networks
presents challenges in achieving real-time translation and localizing the technology for practical use. To address these
challenges, this paper proposes a lightweight sign language recognition network architecture that optimizes the existing
MobileNetV3+LSTM architecture. The goal is to reduce the network model size, decrease computational complexity, and
maintain accuracy. The proposed model was evaluated using a self-collected dataset consisting of 14 types of Taiwanese
daily life sign language gestures. Compared to other models, the proposed model achieved a 99.2% accuracy rate while
reducing complexity by 72%.
Keywords - Sign Language Recognition, Taiwanese Daily Life Sign Language, Lightweight, Edge Device.