Abstract:
The study of football is increasingly taking up a larger portion of contemporary sports intelligence. It assists in obtaining useful information on videos of games to teams, analysts, and researchers. When you do traditional manual analysis, it is difficult to understand how players behave, how tactics are worked out, and how the game may be changed in real time due to the long time of analysis, the ease of error and the impossibility to scale up. The paper outlines a fully computer vision-based system of automatic football match analysis. It involves object recognition, tracking, motion compensation of the camera and the process of assigning players to teams and mining metrics. The framework employs a detection framework based on Yolo to locate the ball, players and referees in video frames. Subsequently, powerful tracking system monitors item IDs with time and estimation of camera movement stabilises the analysis process by compensating the effect of panning and zooming. The system also possesses a view-transformation module which represents a top-down view of the field. This gives you the ability to quantify the speed, range covered by players, the ball action and relative positions of the players. The algorithm is based on color and spatial hints to identify rivals and determine how to assault them in the most favorable manner. Tests indicate that the given framework is capable of detecting and tracking objects specifically and continuing to do so even when the lighting, camera angles, and the number of players vary. Quantitative findings indicate that, stability tracking has been enhanced, identity transition has reduced and metric extraction is now predictable and therefore easier to perform phase-wise tactical analysis. This suggested system proves to be superior and is more objective to perform manual annotation. It also provides you with helpful features to determine your progress, coaching decision making, match intelligence, and future sports analytics.
