Improving Object Detection in Sports Using Deep Learning
Author Information
Author(s): Pei Baocheng, Sun Yanan, Fu Yebiao, Ren Ting
Primary Institution: School of Physical Education, Jinjiang College, Sichuan University
Hypothesis
Can a new supervised object detection method improve accuracy in motion management scenarios for athletes?
Conclusion
The proposed method significantly enhances target detection accuracy in sports scenarios compared to existing networks.
Supporting Evidence
- The proposed method achieved a map_0.5 score of 92.298%, outperforming seven common target detection networks.
- Ablation experiments showed that removing key components reduced detection accuracy.
- The method effectively addresses challenges like temporal information loss and target overlap in sports scenes.
Takeaway
This study created a new way to help computers recognize athletes and sports equipment better during games and training.
Methodology
The study designed a supervised object detection network incorporating a Temporal Shift Module (TSM), a Deformable Attention Transformer (DAT), and a decoupled detection head.
Limitations
The study primarily focuses on sports scenarios and may not generalize to other types of object detection tasks.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website