Abstract
The Hawkeye system was regarded as a successful and effective referee assistant in commercial applications. However, the high hardware investment and strong professionalism set up a physical hurdle in public application. In this sense, one AI referee assistance system is a small-scale system developed under the guidance of computer vision theory and technology, using deep learning-based object detection algorithms and computer graphics-related knowledge. Its main function is to use computer vision technology to make correct and fair judgments on controversial decisions in badminton matches. The AI referee assistance system uses the Harmony OS platform to build a frontend mini-program and uses the YOLOV5 algorithm to detect and track corresponding targets, ultimately achieving judgments both inside and outside the bounds as well as serving violations. In this study, it was found that the object detection model trained in this study performed well, with a prediction accuracy of up to 99%. For posture recognition, this study also utilizes the relatively mature MediaPipe algorithm, which ensures high accuracy of the entire intelligent referee system's ruling and meets the needs of current students. This helps college students better participate in badminton training and competitions, further improving their enthusiasm for participating in sports and, to some extent, solving the problem of scarce referee resources.