Sign Language Translation using Singleshot Mobilenet V2
Asha G Hagargund, Atish Maragur, Abhishek Shetty, A R Bhuvan, Prabhuling · 2024
The use of sign language is essential for deaf and hard-of-hearing individuals to communicate with the world. Traditional methods involving sensors and image processing techniques like edge detection and Hough Transform are often expensive and complex. This research tackles real-time fingerspelling recognition in sign language through advanced machine-learning techniques. A dataset of five distinct gestures was created using webcam images, which the system processes to display corresponding characters on a screen in real time. Leveraging a pre-trained SSD MobileNet V2 model for transfer learning, the system achieves robust and consistent classification of sign language gestures, with an accuracy of over $80 \%$ under various lighting conditions and backgrounds. This approach bridges the communication gap for deaf individuals, enabling effective interaction through hand gestures. The system is user-friendly and scalable, suitable for educational tools, assistive technologies, and real-time communication aids. Future work will expand the gesture dataset, enhance accuracy, and improve robustness in diverse environments, including integrating more complex sign language gestures and phrases to further enhance the system’s utility and performance.