Transformer-Based Multi-Player Tracking and Skill Recognition Framework for Volleyball Analytics
Volleyball is a dynamic sport requiring high technical skills and tactical awareness, demanding effective training methods for performance improvement. Traditional training approaches often rely heavily on subjective analysis by coaches, leading to inconsistencies in skill development. The integrati...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10830493/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Volleyball is a dynamic sport requiring high technical skills and tactical awareness, demanding effective training methods for performance improvement. Traditional training approaches often rely heavily on subjective analysis by coaches, leading to inconsistencies in skill development. The integration of advanced computing technologies has opened new avenues in sports analytics, enabling data-driven methods for performance monitoring and strategy development. Despite advancements in sports like baseball, soccer, and basketball, volleyball has not yet been thoroughly explored for computer-assisted analysis, particularly in player tracking and skill recognition. This study proposes a novel Volleyball Player Tracking and Skill Analytics (VPTSA) framework designed specifically for comprehensive volleyball match analysis. The framework integrates multi-player tracking and action recognition systems, leveraging YOLOv7 for player detection and a memory transformer to maintain spatiotemporal information for accurate tracking. Additionally, a transformer-based action recognition module identifies volleyball maneuvers, providing detailed analytics on player performance and behavior. The proposed methodology effectively overcomes challenges such as occlusions, similar appearances among players, and complex motion patterns, which are prevalent in high-intensity sports like volleyball. Results demonstrate that our model achieves superior performance, with an IDF1 score of 72.6, MOTA of 94.8, and HOTA of 73.7, outperforming state-of-the-art models such as TransTrack and ByteTrack on the SportsMOT dataset. In terms of action recognition, our Volleyball Skill Analytics Network (VSAN) model outperforms existing methods, achieving an individual action mAP of 85.5% and a group action mAP of 90.7% on the Volleyball dataset, demonstrating its efficacy in accurately identifying and classifying volleyball maneuvers. |
|---|---|
| ISSN: | 2169-3536 |