https://ieeexplore.ieee.org/document/10934234
Suruchi Dedgaonkar; Pravin Futane; Ratnamala Bhimanpallewar; Pratham Dedgaonkar; Arjun Deokule; Amodini Dhadge
The article aims to develop and demonstrate an analysis system that can be used for real time analysis in a football match and uses artificial intelligence and computer vision to accurately detect, track, and analyse players and the ball during live matches. The overall goal of it is to transform traditional, intuition-based football analysis into a data driven process allowing coaches, analysts, and broadcasters to make instant tactical and strategic decisions based on live visual and statistical feedback. It highlights how it is being used currently but also what the future may hold for in match analysis
As artificial intelligences involvement in football analytics has evolved rapidly over the past 10 year, it has allowed teams to gather and process large amounts of tracking, event, and biometric data in real time, identifying patterns and insights that are often invisible to the human eye.
This article introduced a system combining computer vision (YOLOv5 and OpenCV), tracking algorithms (SORT and Kalman filters) and data preprocessing techniques to generate precise, real-time insights on player performance, positioning, ball possession, and match dynamics.
One of the studies in this article showed how using model development (YOLOv5) for object detection of players, referees and the ball. The Sort and Kalman filters allows smooth real-time tracking even during occlusion or lighting changes on the pitch and the use of colour segmentation and spatial clustering determine team identification and ball possession.
They also conducted a study which looked at performance evaluation and how they could achieve high accuracy and real-time responsiveness through quantitive measures such as Precision, Recall, Mean Average Precision. This evaluation showed robustness under varying lighting and crowd conditions.
The key findings of their studies showed accurate player and ball detection using YOLOv5 to successfully detect and label all relevant objects in real time as well as using SORT to ensure continuous effective object tracking.
Other outcomes demonstrated how real-world motion data (e.g., 14.8 km/h speed, 19.4 m covered) can help enhance performance assessment of players speed and distance covered and the ability to anaylse the teams possession.
Real-Time Tactical Analysis:
Enables coaches to adjust formations, substitutions, or pressing intensity mid-match based on live analytics.
Player Development and Training:
Provides metrics on speed, endurance, and positioning to support post-match reviews and training adjustments.
Broadcasting and Fan Engagement:
Real-time overlays and visualisations enrich commentary and viewer experience with live statistics.