A Survey on Recognition and Translation System of Real-Time Sign Language

Real-time sign language recognition systems are developed to improve access towards communication and enhance the accessibility of the communication pathway in the community of deaf and hard of hearing. This paper aims to provide a review of various methodologies that have been exploited to interpret such systems, with a focus on advanced deep learning techniques being implemented. The LSTM model applied towards understanding sign language gestures in real time where users can seamlessly communicate using a friendly interface, while capturing gestures through feeds from cameras demonstrates high accuracy in recognizing a vast range of gestures that show potential usage as assistive technology. More recently, CNNs have been extensively used in American Sign Language recognition, and promising results have been achieved with models such as GoogLeNet and YOLO. Relevant models to this paper utilize datasets with widespread lighting conditions and varied backgrounds, creating a more robust model. More recent work has included the YOLOv4-CSP algorithm that produces improved detection performance with optimized network architecture and data augmentation techniques. In addition, YOLO-v9 goes a significant step beyond others in terms of real-time sign language detection. In addition, it is faster in recognizing gestures and has much higher accuracy than any previous version. Such technologies make communication instantaneous but also empower hearingimpaired people by bringing them closer to others in society. Further studies and developments in this area may even lead to more refined solutions for solving sign language translation issues to increase appreciation towards each other.