Badminton trajectory tracking method based on binocular stereo vision and YOLOv8

Badminton trajectory tracking has important applications in sports training analysis and intelligent companion systems. However, current methods face several challenges. High-speed motion makes it difficult for traditional algorithms to capture continuous trajectories. The small target size leads to missed detections during long-distance tracking due to feature loss. Complex environments reduce tracking stability. This paper puts forward a badminton trajectory tracking model based on Binocular Stereo Vision and an improved You Only Look Once version 8 algorithm. The model uses the improved Retina Cortex Theory to enhance the image contrast, applies a matching algorithm to optimize binocular disparity for higher depth accuracy, and integrates a You Only Look Once version 8 algorithm with a Convolutional Block Attention Module for stronger target detection. Experimental results show that the attention-based You Only Look Once version 8 algorithm achieves an average precision of 96.22%, with 97.61% for small targets and 98.64% for medium targets, while the mean absolute error is only 1.26%. The proposed model achieved a tracking success rate of 98.44% and a multi-object tracking accuracy of 96.94% on badminton trajectory and action recognition datasets, with a processing frame rate of 60 and a trajectory breakage rate of only 1.38%. These results outperform traditional models significantly. This research provides technical support for high-precision badminton trajectory tracking and promotes the application of computer vision in intelligent sports systems.