Real-time analysis of soccer ball-player interactions using graph convolutional networks for enhanced game insights.

We present a sequential fusion-based real-time soccer video analytics approach designed to comprehensively understand ball-player interactions. Our approach leverages the power of deep computer vision models, employing a CSPDarknet53 backbone for detection and a Graph Convolutional Network (GCN) for predictive analytics. The proposed approach intricately analyzes ball-player interactions by evaluating metrics such as inter-player distances, proximity to the ball, and hierarchical sorting based on shortest distances to the ball. We also track and estimate each player's total distance and speed covered throughout the game. Our method performs exceptionally well on both uni- and multi-directional player movements, uncovering unique patterns in soccer videos. Extensive experimental evaluations demonstrate the effectiveness of our approach, achieving 91% object detection accuracy, 90% tracking and action recognition accuracy, and 92% speed analysis accuracy on benchmark datasets. Furthermore, our approach outperforms existing GCN techniques, achieving accuracies of 92% in graph connectivity, 89% in node classification, 87% in player tracking, and 88% in event recognition. Here, we show that our method provides a robust and accurate solution for real-time soccer video analytics, offering valuable insights into player performance and team strategies.