Deep Learning Redefines Player Detection: Advancing Sports Performance Analytics

This paper presents a cutting-edge solution for identifying and tracking players in sports broadcast videos an enduring challenge within the Deep Learning and Computer Vision domain. Our approach utilizes a fully autonomous system to extract positional data from soccer match recordings seamlessly integrating components like field registration, player detection, and team assignment. We present an approach using fine-tuned YOLOv8 and Deep SORT for player identification and tracking, assigning unique IDs and enabling automatic statistics generation. Achieving a mean average precision (mAP) of 96%, our model addresses the challenges of tracking players in broadcast soccer videos, especially during rapid movements and difficult visuals. Additionally, we introduced a feature for identifying and tracking the referee and ball. Using Multi-Object Tracking Accuracy (MOTA), our model achieved 96% accuracy in testing.