⚽ Sports Text Commentary Generation using Computer Vision

This project generates automatic football match text commentary using Computer Vision and Deep Learning.
The system detects players and football from a video, tracks their movement, analyzes ball possession, and generates real-time commentary similar to a sports commentator.

The project uses YOLOv8 for object detection, tracking algorithms, and rule-based commentary generation.

🚀 Features

• Detect players and football in match video
• Fine-tuned YOLOv8 model for accurate football detection
• Pretrained YOLO model for player detection
• Object tracking of players and ball
• Ball possession detection
• Automatic sports commentary generation
• Display processed video with commentary
• Web interface using Flask

---

🧠 Project Workflow

The system follows these steps:

1. Input Video

A football match video is provided as input.

2. Object Detection

• YOLOv8 pretrained model detects:
  • Players
• Fine-tuned YOLOv8 model detects:
  • Football

3. Object Tracking

Players and ball are tracked across frames.

4. Possession Detection

Distance between player and ball is calculated.
The closest player is considered the ball owner.

5. Commentary Generation

Based on ball movement and player possession.

Example commentary:

Player 3 passes the ball to Player 7
Player 5 gains possession of the ball
Player 2 intercepts the pass

6. Web Visualization

Processed video and live commentary are displayed on a webpage.

---

🏗️ Project Structure

sports-commentary-ai
│
├── app.py                # Flask web application
├── detector.py           # YOLO detection logic
├── tracker.py            # Object tracking
├── possession.py         # Ball possession detection
├── commentary.py         # Commentary generation logic
│
├── templates
│   └── index.html        # Web interface
│
├── input.mp4             # Input football video
├── output.mp4            # Processed output video
│
├── runs                  # Training results (YOLO)
├── models
│   └── football.pt       # Fine-tuned football detection model
│
└── README.md

---

⚙️ Technologies Used

• Python
• OpenCV
• PyTorch
• Ultralytics YOLOv8
• Flask
• HTML / CSS / JavaScript

---

📦 Installation

Install dependencies:

pip install -r requirements.txt

Or install manually:

pip install ultralytics opencv-python flask torch

---

▶️ Running the Project

Run the Flask server:

python app.py

Open your browser and go to:

http://127.0.0.1:5000

The webpage will display:

• Processed football video
• Generated sports commentary

---

🎯 Model Training (Football Detection)

The football detection model was fine-tuned using a custom dataset.

Training command:

yolo detect train \
model=yolov8n.pt \
data=data.yaml \
epochs=50 \
imgsz=640

This improves detection accuracy for small football objects.

---

🧪 Example Commentary Output

Player 4 gains possession of the ball.
Player 4 passes the ball to Player 8.
Player 8 moves forward with the ball.
Player 10 intercepts the pass.

---

📊 Future Improvements

• Real player identification
• Team detection
• Goal detection
• Advanced NLP-based commentary
• Audio commentary generation
• Live match streaming support

---

👨‍💻 Author

Love Kumar
B.Tech Computer Science
Lovely Professional University