Shooting Prediction Based on Vision Sensors and Trajectory Learning

Abstract

Basketball has become one of the most popular sports and is generally popular in international sports events. However, how to effectively achieve shooting prediction and then guide shooting has become a major challenge. Different from the classical manual observation method, this paper proposes a real-time shooting prediction method based on vision sensors and trajectory learning. In the research, we first extracted the basketball trajectory information on template matching and centroid calculation and then obtained a smooth trajectory curve through interpolation. Taking the change of x, y coordinate position, height Y, and distance D from the shooting point during the instantaneous movement as basic features, four machine learning algorithms were used to analyze the impact of different feature combinations on the shooting prediction. Finally, we analyzed the minimum trajectory point requirements predicted when making a shot. The experimental results show that our method can effectively predict the effect of shooting when the feature combination is basketball height and time. When the interpolation density is high (the total number of trajectory points is 116), the overall accuracy can reach more than 90%, and only one-third of the effective trajectory length is required, which effectively helps athletes improve their shooting percentage and assist referees in daily training.