A badminton serve violation detection model based on adaptive addition variant algorithm

Abstract

The detection of badminton serve violations is of great significance for improving the fairness of the game and the efficiency of refereeing. However, the complex sports scenes and changing lighting conditions make it difficult for traditional detection methods to meet real-time and accuracy requirements. Therefore, the study proposes an object detection model based on the improved “You Only Look Once” (YOLO) variant algorithm and a serving state discrimination model based on angle quantification analysis. First, a multi-layer feature adaptive fusion module is utilized to achieve high-precision real-time positioning of the shuttlecock and racket. Second, the dynamic features of the racket and target point are mapped into computable geometric quantities, and a state discrimination model is designed. The performance test results of the object detection model show that compared to the original model, its average accuracy, recall rate, and precision rate increased by 1.85%, 3.07%, and 4.99%, respectively, and the inference time increased by 4.12%. The light adaptability in low-light environments is only 4.8%. In the application testing of the violation detection model, the false detection rate is 7.8% when the sample size is 5000. When the offset value is 25 px, the accuracy of violation detection is 88.9%, and the processing frame rate is 47.7. The experimental results show that the model exhibits strong robustness under various lighting conditions and complex backgrounds, with better detection accuracy and speed than existing methods. It can provide efficient support for the determination of serve violations in badminton matches.