Efficient tracking of team sport players with few game-specific annotations

Abstract

One of the requirements for team sports analysis is to track and recognize players. Many tracking and reidentification methods have been proposed in the context of video surveillance. They show very convincing results when tested on public datasets such as the MOT challenge. However, the performance of these methods are not as satisfactory when applied to player tracking. Indeed, in addition to moving very quickly and often being occluded, the players wear the same jersey, which makes the task of reidentification very complex. Some recent tracking methods have been developed more specifically for the team sport context. Due to the lack of public data, these methods use private datasets that make impossible a comparison with them. In this paper, we propose a new generic method to track team sport players during a full game thanks to few human annotations collected via a semi-interactive system. Non-ambiguous tracklets and their appearance features are automatically generated with a detection and a reidentification network both pre-trained on public datasets. Then an incremental learning mechanism trains a Transformer to classify identities using few game-specific human annotations. Finally, tracklets are linked by an association algorithm. We demonstrate the efficiency of our approach on a challenging rugby sevens dataset. To overcome the lack of public sports tracking dataset, we publicly release this dataset at https://kalisteo.cea.fr/ index.php/free-resources/. We also show that our method is able to track rugby sevens players during a full match, if they are observable at a minimal resolution, with the annotation of only 6 few seconds length tracklets per player.

Multi-task Learning for Joint Re-identification, Team Affiliation, and Role Classification for Sports Visual Tracking

“…More recently, reidentification-based trackers, like StrongSORT [12], have emerged, which focus on extracting more discriminative features for object re-identification to improve tracking results [2,29,57]. Some works have also been specifically developed for tracking team sport players [16,22,30,32,33]. [55] proposes a deep learning-based approach for multi-camera multi-player tracking in sports videos, leveraging deep player identification to improve tracking accuracy and consistency across multiple cameras.…”

“…Despite substantial advancements in sports analysis methods, as shown in recent studies [8,14,19,27,30,45,53], the majority of current tracking methods do not tackle all these tasks together. Solving each task individually is also not optimal as it overlooks the common objectives shared by all three tasks for accurately representing the individual, which could potentially benefit from a unified approach.…”

Sports competition tactical analysis model of cross-modal transfer learning intelligent robot based on Swin Transformer and CLIP

“…They exhibit high flexibility and scalability (Victor et al, 2021 ), making them suitable for capturing complex and dynamic interactions in various sports competitions. However, the performance of GNNs heavily relies on the completeness and quality of graph data (Maglo et al, 2022 ). Without a clear, complete, and accurate graphical representation, the model may fail to capture key inter-entity relationships, which may make it difficult for the model to understand the tactical relationships between opposing players.…”

Individual Locating of Soccer Players from a Single Moving View

“…For example, Theagarajan and Bhanusome [ 15 ] use YOLOv2 [ 79 ] to detect soccer players, while Zhang et al [ 80 ] use Mask R-CNN [ 55 ]. Hurault et al [ 16 ], Vats et al [ 81 ], and Maglo et al [ 82 ] use a Faster R-CNN model [ 83 ] to detect soccer, hockey, and rugby players. The YOLOX detector [ 84 ] has also been used in many recent frameworks [ 17 , 18 , 85 ] since it offers a state-of-the-art trade-off between detection speed and accuracy.…”