🏀 NBA Vision360 – Real-Time Basketball Video Intelligence

NBA Vision360 is a real-time AI-based video analytics system that turns raw basketball footage into ESPN-style stat overlays, predictive modeling, and AI-generated commentary. Designed for analysts, fans, and coaches, it combines computer vision, machine learning, and natural language processing into one streamlined pipeline.

---

🎯 Features

👁️ Player Detection & Stat Overlay

• Uses YOLOv5 + OpenCV to detect players and overlay circular stat rings
• Highlights player metrics like PPG, PER, 3P%, and more

📊 Win Probability Prediction

• Computes win probabilities dynamically using logistic regression
• Inputs include: quarter, time left, and score, parsed from live OCR

🧮 Player Performance Rating

• Calculates game-specific ratings using historical stats, live performance, and spread-based modeling
• Includes stamina and fatigue estimation based on pace and minutes

🗣️ AI Commentary System

• Uses Edge-TTS to generate real-time play-by-play commentary
• Triggered by high-leverage moments, clutch plays, or rating spikes

🗺️ Tactical View Conversion

• Converts broadcast camera view to tactical court positioning using homography
• Useful for coaching-level breakdowns

---

🛠️ Tech Stack

Language: Python
Computer Vision & OCR: OpenCV, YOLOv5, pytesseract
ML/Stats: scikit-learn, pandas, NumPy
Audio: Edge-TTS
Data: Custom JSON stat feeds, CSV coefficients
Visualization: OpenCV overlays (stat rings, fatigue metrics)

---

📂 Folder Structure

NBA-Vision360/
├── main.py                         # Entry point
├── predictor.py                   # Win probability engine
├── commentary/                    # Edge-TTS commentary system
├── Rating/                        # Player rating logic
├── court_keypoint_detector/      # Homography & tactical conversion
├── score_detector/               # Scoreboard OCR
├── team_assigner/, trackers/     # Player tracking logic
├── heat_map_players/, speed_and_distance_calculator/
├── pass_and_interception_detector/, ball_acquisition/
├── input_videos/                 # (sample game clips)
├── requirements.txt
├── README.md

---

---

📌 Roadmap

• Win probability + rating model
• TTS commentary engine
• Scoreboard OCR + quarter tracking
• Assist prediction engine (in progress)
• Real-time Web UI or Streamlit demo
• Support for multiple camera angles

---

📄 License

This project is licensed under the MIT License. See the LICENSE file for full details.

---

🙌 Acknowledgements

• ESPN – for inspiring the analytics experience
• Roboflow + YOLOv5 – for model training support and object detection
• NBA stats datasets – cleaned and compiled from public sources
• abdullahtarek/basketball_analysis – for foundational ideas and structure around basketball stat overlays and player detection

---