Research on Volleyball Players Tracking Based on Improved DeepSORT

Abstract

In order to reduce the problems of identity switching caused by occlusion in volleyball video, we propose a method named Vol-DeepSORT based on DeepSORT for tracking volleyball players. According to the principle that the number of players is unchanged in a fixed playing area, we improved the DeepSORT algorithm that rematches the new tracks with the vanishing tracks by the minimum Euclidean Distance, so as to realize the recovery of the player identity. We use the improved algorithm Vol-DeepSORT combined with YOLOv5 to detect and track the players. Experimental results show that the MOTA of Vol-DeepSORT is 90.07%, and the frequency of identity switching is reduced by 30% compared with the original DeepSORT. The proposed algorithm can reduce the frequency of identity switching on players to a certain extent and achieve a good tracking performance.