Badminton Detection Using Lightweight Neural Networks for Service Fault Judgement

The Badminton World Federation stipulates that the height of the badminton during the serving shall not exceed 1.15 m. In current official matches, enforcement still relies on the referee’s eye judgment. The Hawk-Eye system in matches mainly judges shuttlecock landings, which is not suited for service fault judgement. Recently, some algorithms based on computer vision have been proposed for badminton rules judgement. However, it is a challenging task for badminton service fault detection. Firstly, the ball is usually small and moves fast within the imaging field of view, complicating the detection process. Furthermore, accurately locating the service point and verifying its height against standards is technically challenging. Additionally, the scarcity of public datasets for badminton detection poses further challenges to developing effective detection algorithms. To address these deficiencies, an annotated dataset is collected for badminton detection. A deep learning network based on the concept of CenterNet is proposed to locate badminton in the imaging field of view accurately and in real time. Experiments are conducted on the collected dataset compared with four state-of-the-art methods. The experimental results demonstrate that our method used for badminton detection can achieve high detection accuracy in real-time on ordinary computing devices such as personal computers or smartphones.