Automated Football Detection, Tracking, And Object Detection, And Homography

Abstract

Modern Football Has Transitioned Into A Data-centric Era, Where Tactical Efficiency Is Often Measured Through Granular Metrics Rather Than Just Match Outcomes. This Project Addresses The Gap Between Raw Broadcast Footage And Structured Analytical Data By Constructing A Custom Computer Vision Pipeline. Instead Of Relying On Expensive, Proprietary Tracking Systems Used By Elite Clubs, We Developed A Modular System Capable Of Extracting Player Trajectories, Team Formations, And Physical Performance Indicators From Standard Single-camera Video Feeds. Our Implementation Integrates YOLOv8 For Robust Object Detection With The SORT Algorithm For Real-time Tracking, Enhanced By A Custom Homography-based Camera Motion Compensation Module. A Key Focus Of Our Work Was Mathematically Decoupling The Camera’s Panning And Zooming Movements From The Actual Player Velocity, A Common Source Of Error In Amateur Ana- Lytics Projects. By Mapping Pixel Coordinates To A Real-world Pitch Model, We Successfully Derived Actionable Insights Such As Heatmaps And Sprint Profiles. This Paper Detailed The Specific Engineering Challenges Encountered—from Handling Occlusion In Crowded Penalty Boxes To Calibrating Color Thresholds For Jersey Segmentation—and Presents A Scalable, Open-source Approach To Democratizing Sports Analytics.

Keywords