Low-cost Badminton Trajectory Recognition and Landing Point Prediction Optimization Based on Field Coordinate System Transformation

Judging the flight trajectory curve and predicting the landing point are important factors affecting the quality of hitting in badminton matches. Many badminton professionals and coaches usually analyze the strength of badminton matches through video analysis to improve their own abilities and develop competition strategies. This study proposes a field coordinate transformation strategy for badminton trajectory recognition to reconsider the badminton field, and then proposes a badminton detection tracking and trajectory prediction algorithm based on video data streams. This algorithm is divided into badminton tracking, recognition detection, and trajectory prediction. The results indicated that outdoor environments had better accuracy in anchor coordinate conversion. The offset distance for anchor coordinate conversion was above 0.75m indoors and between 0.25m-0.05m outdoors. The recognition accuracy was over 95%, with a minimum of around 80%, and the algorithm took between 5s to run in different environments. During different stages of operation, the average running time, average coordinate calculation time, and average trajectory prediction time of the algorithm were 621.25ms, 16.75ms, and 342.875ms, with a standard deviation of 12.277, 0.782, and 4.552. The research algorithm was cost-effective in terms of running time and effectiveness. The improved algorithm in this study had good recognition performance with accuracy, recall, F1, and frame rate of 98.6%, 97.6%, 98.7%, and 30.8 frames/s, respectively. The research methods and results have promoting significance for the development of badminton sports.