American Sign Language Detection using YOLOv5 and YOLOv8

Abstract

Abstract In the modern world, culture and religion are diverse and widespread. Sign language culture had grown since its emergence in the American School for the Deaf (ASD) in 1817. In a world where computers are now solving real-time applications and issues using deep learning, Sign language (SL) is one of those. YOLO is an object detection and classification algorithm that uses Convolutional neural network (CNN) to achieve high performance and accuracy. The paper aims to detect American sign language using YOLO models and compare different YOLO algorithms by implementing a custom model for recognizing sign language. The experiments show that the latest YOLOv8 gave better results than other YOLO versions regarding precision and mAP, while YOLOv7 has a higher recall value during testing than YOLOv8. The proposed model is lightweight, fast and uses the American sign language letters dataset for training and testing. The custom model achieved 95% precision, 97% recall, and 96% mAP @0.5, showing the model capabilities in real-time hand gesture recognition.