The study of object tracking has substantially advanced thanks to the development of deep learning visual recognition and tracking methods. However, because to the additional difficulties they provide, such as the difficulty in tracking small, swiftly moving objects like a ball or shuttlecock due to the fast camera movement and the existence of swings and spins, sports videos are still understudied. To access these massive archives of sports video data and automatically tag and analyse its properties, such as player performance and stroke and shot analysis, an effective end-to-end solution is needed. The aim of this research is to create a complete deep learning based model that can do object detection and tracking in sports movies as well as classify the played stroke. We employed the SF-YOLOv5 model, a lightweight model for the identification of swiftly moving small objects, for this. Then, we utilised the Deep-Sort algorithm and zero shot learning to follow the objects that had been detected. Finally, we classified the played shot using the CNN classifier.
© Copyright 2023 Innovation and Technology in Sports: Proceedings of the International Conference on Innovation and Technology in Sports, (ICITS) 2022, Malaysia. Alle Rechte vorbehalten.