Research on Motion Recognition and Real-Time Classification Algorithm of Sports Event Video Based on Deep Learning

This paper designs a supervised learning method for athletes' movements in sports videos. Firstly, the concrete motion track is extracted by means of grid segmentation, and the corresponding motion features are extracted based on the characteristics of light flow in the arena. Then a guided SVM model is used to identify the categories of player behavior. This paper presents a method to solve the motion characteristics of objects by using grid bar graph. For the first time, the optical flow vector of the optical flow field is regarded as a moving factor and a kind of spatial distribution information, which is more robust than the conventional algorithm. Through testing and analysis of the proposed method, the results show that the proposed method can significantly improve the robustness of optical flow characteristics. It can realize the automatic detection of sports personnel behavior. This method can be used as a reference for the subsequent research on behavior recognition.