🎾 Padel AI System

Computer Vision applied to Security and Sports Analysis

---

📖 Introduction

Padel AI System is a comprehensive Computer Vision project developed for the Computer Vision I course at Universidad Pontificia Comillas (ICAI).

This system implements a two-module solution using a single mobile device camera:

1. Security System: An access control module based on geometric pattern recognition that acts as a visual password.
2. Tracking System (AI Chair Umpire): An autonomous sports analysis engine capable of tracking a padel ball, detecting players via pose estimation, and validating serves (Valid/Fault) in real-time.

📂 Repository Structure

The project is organized into modular directories to ensure scalability and order:

├── 📂 assets/                   # Images and visual content, e.g., for the ReadMe.md banner
├── 📂 calibration_process/      # Scripts and images for intrinsic camera calibration
├── 📂 complete_padel_system/    # Unified application integrating Security + Tracker
├── 📂 security_system/          # Independent geometric pattern recognition module
├── 📂 tracking_system/          # Independent ball tracking and umpire logic module
├── 📄 .gitignore                # Git configuration
├── 📄 ReadMe.md                 # Project documentation
└── 📄 documentation.pdf         # Final report and project documentation
└── 📄 requirements.txt          # Dependencies and required libraries

🛠️ Technologies and Methodology

The system relies on a hybrid approach combining Classical Computer Vision and advanced algorithms such as those offered by the YOLO library:

• Core Framework: Python 3, OpenCV (cv2).

• Deep Learning: YOLOv8-Pose (Ultralytics) for player keypoint extraction (waist/feet).

• Classical Techniques:

  • HSV color segmentation and MOG2 background subtraction for ball detection.
  • Optical Flow (Lucas-Kanade) for trajectory smoothing.
  • Kalman Filters for state prediction and occlusion handling.
  • Geometric approximation (Douglas-Peucker) on the Convex Hull for the security module.
  • Use of morphological operations, thresholding, binarization, etc.

• Interface: Streamlit for the web dashboard and real-time visualization.

• Optimization: Multithreading for video capture and optional TensorRT support for inference.

🚀 Installation and Setup

Prerequisites

Ensure you have Python 3.9 or higher installed.

1. Clone the Repository

git clone https://github.com/andresgilvicente/padel-ai-system.git

2. Create Virtual Environment (Recommended)

python -m venv venv # On Windows

.\venv\Scripts\activate # On Mac/Linux

source venv/bin/activate

3. Install Dependencies

All necessary libraries are listed in requirements.txt.

pip install -r requirements.txt

🖥️ Execution

You can run the modules independently or as a complete system.

Option A: Complete System (Security + Tracker)

Executes the full flow. You must pass the security check (show 4 geometric shapes) to unlock the tracker.

streamlit run complete_padel_system/complete_padel_system_app.py

Option B: Tracking System (Umpire Only)

Launches the AI Chair Umpire directly for analysis or debugging.

streamlit run tracking_system/tracking_system_app.py

Option C: Security System

Tests the geometric pattern recognition logic in isolation.

streamlit run security_system/security_system_app.py

📊 Feature Description

🔒 Security Module

• Shape Detection: Identifies Lines, Triangles, Squares, Rectangles, Circles, and Pentagons using contour analysis and convex hulls.
• Sequence Decoder: Unlocks the system only when a specific sequence of 4 predefined geometric shapes is detected.

🎾 Tracking Module (AI Umpire)

• Ball Tracking: Hybrid detection using Color/Motion + Kalman Filter Prediction.

• Serve Validation:

  • Detects serve impact based on acceleration peaks.
  • Compares ball height vs. player waist height (YOLO Keypoints).
  • Classifies the serve as VALID or FAULT.

• Bounce Detection: Analyzes the trajectory on the Y-axis to detect direction changes (bounces) on the court.

• Evidence Generation: Automatically saves "Photo Finish" frames of each analyzed serve.

👥 Authors

• Jorge Carnicero Príncipe
• Andrés Gil Vicente

Completion Date: January 10, 2026