Sign Language Recognition Modeling With Deep Learning Method

Abstract

This study presents the development of a deep learning-based system for the recognition of Turkish Sign Language (TİD) using a lightweight and accessible architecture. The proposed model was trained on a custom dataset consisting of 15 hand gesture classes and implemented through Apple’s Create ML platform, leveraging the ResNet50 convolutional neural network. To extract spatial features from hand gestures, the MediaPipe framework was utilized, enabling robust tracking of 21 hand landmarks. By integrating pre-trained models and simplified feature representations, the system achieves a balance between classification accuracy and energy efficiency. Experimental results show that the model reached an overall classification accuracy of approximately 94%, demonstrating strong performance even under variable lighting and background conditions. Additionally, a confusion matrix analysis highlights the system’s strengths and reveals certain class-specific ambiguities, particularly among visually similar gestures. The model’s lightweight structure and minimal hardware requirements make it suitable for deployment in educational institutions, healthcare facilities, and other public domains. This study contributes to the growing body of research on inclusive communication technologies and offers a practical step toward real-time, device-level sign language recognition in underrepresented languages such as TİD.