Real Time Sign Language Recognition Using Yolov5
Dhruv Biyani, Nitika Vats Doohan, Manas Rode, Darshan Jain
Abstract
This paper proposes real-time sign language detection models for deaf and mute individuals created using the You Only Look Once version 5 (YOLOv5) algorithm. There were 3 different models created. The first model detects American Sign Language, the Second model detects Hindi/ Marathi Sign Language and the third model detects static gestures. The YOLOv5 models were trained on the custom data set, achieving high accuracy. The model's real-time performance was evaluated by capturing live sign language gestures using a webcam, achieving an average detection time of 0.05 seconds per frame. This demonstrates the model's potential for real-time sign language interpretation and communication for deaf and mute individuals. To implement the models, our custom data set includes a total of 1892 images, consisting of 452 training images of static gestures, 611 training images of Hindi/Marathi sign language, and 829 training images for American sign language representing different sign language gestures. The proposed models performed better than other deep learning models, including Faster R-CNN and Mask R-CNN, showing superior accuracy and real-time performance. The results demonstrate the potential of using YOLOv5-based models for real-time sign language detection and interpretation, which can significantly improve the quality of life for deaf and mute individuals.