With the development of object detection and re-identification algorithms, multi-object tracking (MOT) has made rapid progress. However, tracking multiple athletes with similar appearances and nonlinear movements in team sports remains a pressing challenge. Current motion-based tracking algorithms typically employ the Kalman Filter to predict target motion, but they struggle to handle nonlinear movements and mutual occlusions among multiple targets effectively. To address this, we propose a MOT framework based on improved OC-SORT and motion information estimation, utilizing the Transformer architecture as a motion predictor instead of the Kalman Filter. We introduce historical trajectory embeddings to extract spatiotemporal features from sequences of previously detected bounding boxes. During the association phase between current detections and historical trajectories, accurate matching is achieved u-sing the Hungarian algorithm based on Buffered Intersection Over Union (BIoU). Furthermore, we design different post-processing procedures for three team sports, basketball, soccer, and volleyball, to effectively handle athletes' nonlinear movements on the field. Experimental results show that the proposed method achieves HOTA score of 77. 3% and IDF1 of 78. 2% on the SportsMOT public dataset, outperforming other state-of-the-art methods. This demonstrates the effectiveness and robustness of the proposed framework in MOT tasks for various team sports，including basketball，soccer，and volleyball.