In the multi-obect tracking (MOT) algorithm for sports scenes, the motion model based on Kalman filter cannot cope with the complex motion of objects and the frequent entry and exit of players in videos, which inevitably leads to identity switches. To solve this problem, we propose a bidirectional buffer trajectory management algorithm, based on the prior information that the number of players on the sports field will not change easily, we propose a bidirectional trajectory buffer method. When initializing a new trajectory, the new trajectory needs to match with the previously lost trajectory in advance, and try to recover the trajectory that once left the image, reducing the identity switch. We conducted comparative experiments on SportMOT and DanceTrack to prove the effectiveness of the proposed algorithm. Compared with the benchmark algorithm, the method improved HOTA by 1.1% on SportsMOT and reduced IDSw by 9.5%. On DanceTrack, HOTA was improved by 0.5%, and IDSw was reduced by 14%.