Research and Improvement of Chinese Sign Language Detection Algorithm Based on YOLOv5s

Abstract

In order to facilitate the daily communication of hearing-impaired people, a model based on target detection for real-time recognition of sign language is studied to allow ordinary people who do not know sign language to communicate fluently with hearing-impaired people. YOLOv5 target detection network model has a series of advantages such as fast and efficient detection and easy deployment. This paper improves yolov5s network model applied to sign language recognition detection by adding an attention mechanism module to spatially and channel to extract effective information. And improve the loss function to improve the regression progress. The model is trained with a homemade sign language dataset, and the trained model is used to detect and recognize sign language and convert between sign language, speech and text. In tests in real-world environments, the improved yolov5s-based sign language network detection model can quickly and effectively recognize more than forty basic everyday sign languages with a detection accuracy of 96.50% and an average detection accuracy mAP of 98.92%. The results show that the improved yolov5s-based sign language detection model can accurately and effectively recognize sign language gestures in different environments, and its easy deployment in mobile devices can help hearing impaired people in their daily communication.