An Innovative Approach for Detecting and Visualizing the Trajectory of Curling Stones in Competition Broadcast Videos

The broadcast videos of curling competitions contain a large amount of professional information. Extracting the trajectories of curling from these videos is of great significance for improving the competition and scientific research levels in the field of curling. However, due to the interference of factors such as dynamic viewing angles, local fields of view, and frequent occlusions, trajectory detection faces many challenges. To this end, in view of the characteristics of curling sports, this paper proposes an innovative method to accurately detect and extract the movement trajectories of curling stones in competition broadcast videos. This method first combines YOLOX and the Hough transform to detect the pixel positions and angles of curling stones in each video frame. Then, a feature point detection method based on semantic segmentation is used to obtain the projection transformation matrix of each video frame, and the detection results are converted into physical coordinates. Finally, by designing an innovative matching strategy, an improved SORT algorithm is constructed to achieve the tracking of curling stone targets across video frames and the construction of smooth trajectories. The experimental results in the broadcast videos of the Winter Olympics show that the proposed method outperforms the existing mainstream algorithms in multiple performance indicators. Especially in terms of trajectory detection accuracy, it is 21.14% higher than the baseline model. This method can not only provide reliable technical support for the real-time analysis of curling competitions but also provide an important data foundation for the research in fields such as curling robots.