Basketball-SORT: An Association Method for Complex Multi-object Occlusion Problems in Basketball Multi-object Tracking

TL;DR

Basketball-SORT tackles Complex Multi-object Occlusion (CMOO) in basketball multi-target tracking by shifting from IoU-based to trajectory-based association on a projected court plane. It introduces Basketball Game Restriction (BGR) to fix the number of active identities and Reacquiring Long-Lost IDs (RLLI) to rebind IDs after long occlusions, while employing occlusion-aware strategies that differentiate same-team and different-team interactions (STO/DTO) using appearance and motion cues. The method achieves state-of-the-art performance on a basketball fixed-camera dataset, with a HOTA of $63.48\%$ and notable gains in association accuracy compared to baselines, demonstrating robust online tracking under CMOO. This approach advances sports MOT by enabling reliable, real-time player trajectories in occlusion-rich basketball scenes, with potential implications for performance analysis and analytics.

Abstract

Recent deep learning-based object detection approaches have led to significant progress in multi-object tracking (MOT) algorithms. The current MOT methods mainly focus on pedestrian or vehicle scenes, but basketball sports scenes are usually accompanied by three or more object occlusion problems with similar appearances and high-intensity complex motions, which we call complex multi-object occlusion (CMOO). Here, we propose an online and robust MOT approach, named Basketball-SORT, which focuses on the CMOO problems in basketball videos. To overcome the CMOO problem, instead of using the intersection-over-union-based (IoU-based) approach, we use the trajectories of neighboring frames based on the projected positions of the players. Our method designs the basketball game restriction (BGR) and reacquiring Long-Lost IDs (RLLI) based on the characteristics of basketball scenes, and we also solve the occlusion problem based on the player trajectories and appearance features. Experimental results show that our method achieves a Higher Order Tracking Accuracy (HOTA) score of 63.48$\%$ on the basketball fixed video dataset and outperforms other recent popular approaches. Overall, our approach solved the CMOO problem more effectively than recent MOT algorithms.

Figures (4)

• Figure 1: An example of Complex Multi-object Occlusion (CMOO) problems with a sequence of (a) to (c), including three or more player occlusions with similar appearance of players' jerseys in the same team and high-intensity complex and unpredictable motion. This can greatly affect detection and tracking accuracy and will cause a player to be lost and an ID switch.
• Figure 2: The pipeline of our Basketball-SORT algorithm. The main technical contributions are BGR and RLLI. BGR: In a basketball game, there can only be 10-player motion trajectories. At the 100th frame, we calculate the 10 longest trajectories on the court and identify them as players. RLLI: For the Long-Lost state player, we compare the position and appearance information of the Long-Lost player with the new detection bounding box to determine whether the player tracking ID should be reacquired.
• Figure 3: Standard basketball court projection, We use the red dots in the figure above to convert the image coordinate system coordinates to the player's position in the basketball court coordinate system
• Figure 4: (a-c) show the results using the DeepEIoU method. During occlusion, the player ID is lost, and the method fails to assign the player ID correctly. (d-f) demonstrate the results using the Basketball-SORT approach, which successfully matches the player ID during complete occlusion.