⚽ Football Player Tracking & Analytics System (YOLO + CV + AI)

This project detects and tracks football players, referees, and the ball from broadcast match footage using YOLO Object Detection and advanced Computer Vision techniques.
It assigns players to teams using KMeans color clustering, calculates ball possession percentage, estimates speed & distance covered, and stabilizes tracking even under camera movement using optical flow compensation.

---

🎥 Demo

Input Match Footage	Output With Tracking & Analytics
Raw broadcast video	Real-time tracking & possession stats

Demo GIF Preview 👇

---

🚀 Features

Module	Description
YOLOv8 Object Detection	Detects players, referees, and ball
ByteTrack Tracking	Stable multi-object ID tracking
KMeans Clustering	Team assignment based on jersey color
Ball Possession Analytics	Calculates team ball control percentage
Optical Flow Camera Stabilization	Compensates for panning camera
Perspective Transformation	Field view correction to measure in meters
Speed & Distance Estimation	Compute movement metrics per player

---

📂 Project Structure

├── assets/
│   └── demo.gif
├── input_videos/
│   └── 08fd33_4.mp4
├── output_videos/
│   └── output_video.avi
├── models/
│   ├── best.pt
│   └── last.pt
├── stubs/
│   ├── track_stubs.pkl
│   └── camera_movement_stub.pkl
│
├── trackers/
│   └── tracker.py
├── team_assigner/
│   └── team_assigner.py
├── player_ball_assigner/
│   └── player_ball_assigner.py
├── camera_movement_estimator/
│   └── camera_movement_estimator.py
├── view_transformer/
│   └── view_transformer.py
├── speed_and_distance_estimator/
│   └── speed_and_distance_estimator.py
│
├── utils/
│   ├── bbox_utils.py
│   └── video_utils.py
│
├── training/
│   └── football_training_yolo_v5.ipynb
│
├── metrics/
│   └── results.png
├── requirements.txt
├── main.py
└── README.md

---

🧠 How It Works (Pipeline Architecture)

flowchart LR
A[Input Video] --> B[YOLO Detection]
B --> C[ByteTrack Tracking]
C --> D[Camera Movement Estimation]
D --> E[Perspective Transform]
E --> F[Speed & Distance Calculation]
C --> G[KMeans Team Assignment]
C --> H[Ball Assignment]
H --> I[Ball Possession %]
F --> J[Annotated Output Video]

Loading

---

📦 Installation

git clone https://github.com/yourname/football-tracking-analytics.git
cd football-tracking-analytics
pip install -r requirements.txt

---

▶️ Running the Project

python main.py

Output saved to:

output_videos/output_video.avi

---

🏋️ Model Training (YOLO)

Dataset structure used for training YOLO:

datasets/
  ├── images/train
  ├── images/val
  ├── labels/train
  └── labels/val

Example command:

yolo train model=yolov8x.pt data=football-dataset.yaml imgsz=640 epochs=100 batch=16

---

📈 Results & Metrics

Metric	Value
mAP@50	0.92
mAP@50-95	0.78
Precision	0.89
Recall	0.87

Detection Performance Visualization

---

🛠 Tech Stack

• Python, OpenCV, Numpy, Pandas
• YOLOv8 (Ultralytics)
• Supervision (ByteTrack)
• Scikit-learn (KMeans)
• Optical Flow (Farneback)
• Perspective Transform (Homography)

---

🚀 Future Improvements

• Jersey number recognition
• Automated event detection (pass, foul, shot)
• Predict xG & tactical heatmaps
• Real-time live stream integration

---

📧 Contact

👤 Priyangshu Majumder 📍 India 💼 Email: priyangshumajumder9@gmail.com

---