← sources

I built an end-to-end AI football analytics system using YOLO11 and BoT-SORT

source: reddit (reddit:computervision)published 2026-10-09 17:14ingested 2026-10-10 02:35

retained text

Hey everyone!

I've been working on a computer vision project that turns regular football match footage into player tracking, performance statistics, and tactical insights.

It started as my graduation project, but I wanted to go beyond basic object detection and build a complete video analytics pipeline.

Here's what the system does:

\- Detects players, goalkeepers, referees, and the ball using a custom-trained YOLO11m model.  
\- Tracks players using BoT-SORT, with additional logic to handle broken tracks and ID switches.  
\- Automatically identifies teams based on jersey colors.  
\- Uses pitch keypoint detection and homography to map player positions onto a tactical minimap.  
\- Estimates player speed and distance covered.  
\- Generates heatmaps, possession statistics, tactical insights, and match reports.

**Tech stack :**Python, YOLO11, BoT-SORT, OpenCV, NumPy, and PyTorch.

One of the biggest challenges has been maintaining consistent player identities, especially during occlusions, camera movement, and crowded situations.

I'm continuing to improve tracking reliability and overall robustness.

I'd really appreciate feedback from anyone working in computer vision or sports analytics.

**A few questions:**  
\- What approaches would you recommend for reducing ID switches during heavy occlusions?  
\- How would you improve tracking consistency across changing camera views?  
\- What would you prioritize before using a system like this in production?

GitHub: https://github.com/Mhmdwael77/ai-football-analytics

Thanks for checking it out! ⚽

document fetched 2026-10-10 00:23 · canonical https://reddit.com/r/computervision/comments/1x1r3ff/i_built_an_endtoend_ai_football_analytics_system