🏀 CourtVision: An AI-Enhanced Basketball Score Counter

CourtVision is an AI-enhanced basketball score counter using YOLO for object detection and SORT for tracking. It detects the ball, players, and rim to automatically count the scores for two teams.

📋 Requirements

• Python 3.x
• PyTorch 2.3.1-CUDA:12.1
• Ultralytics
• OpenCV
• YOLOv10
• SORT algorithm: https://github.com/abewley/sort
• SciPy
• NumPy
• collections (defaultdict)

🛠️ Installation

1. Clone the repository:

   git clone https://github.com/AJ-C22/Basketball-ComputerVision.git
   cd Basketball-ComputerVision

2. Install the dependencies:

   pip install -r requirements.txt

3. Download the YOLO model:

   Ensure you have a trained YOLO model named best.pt in the project directory.

🚀 Usage

1. Run the script:

   python hoop.py

2. Control Instructions:

   • Mouse Click: 🖱️ Assigns a player to team 2 (Orange) or toggles back to team 1 (Blue).
   • Press 'q': ❌ Quit the application.

🌟 Features

• Object Detection: Utilizes YOLO for detecting the ball, players, and rim.
• Object Tracking: Uses DeepSORT and Centroid Tracking to track players and ball movements.
• Scoring Mechanism:
  • Detects when the ball passes through the rim to count scores.
  • Maintains separate scores for two teams (Orange and Blue).
• Interactive Team Assignment:
  • Click near a player to assign them to a team.
  • Team 1 (Blue) and Team 2 (Orange) can be distinguished by bounding box colors.
• Shooting Zone Detection:
  • Identifies the shooting zone for each player.
  • Changes color to indicate the ball's presence in the shooting zone.

💡 Implementation Details

YOLO Model

• The YOLO model is used for object detection with a custom-trained model best.pt.
• Classes detected include ball, person, rim, and threepoint.

Scoring Logic

• The ball's position relative to the rim is used to determine scores.
• Separate counters for each team track the number of successful shots.

Dots for Ball Trajectory

• Gaussian filtering smooths the ball's trajectory for better visualization.
• The ball's trajectory above the rim is visualized using green dots.

🔧 Customization

• Frame Skip: Adjust the self.frame_skip variable to process every nth frame for performance optimization.
• Confidence Threshold: Modify the conf > 0.4 line to change the detection confidence threshold.
• Box Sizes: Change box_width and box_height for the rim boxes to adjust the size of the top and bottom scoring zones.

📌 Notes

• Ensure your webcam is properly connected for real-time video capture.
• The performance may vary based on the hardware and quality of the YOLO model used.

📜 License

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

🙏 Acknowledgments

• YOLO
• DeepSORT

📋 To-do

• add a react frontend to make program easily accessible
• add voice commands (read score, remove last point, set score, new game)
• Add 3 point-line
• implement position estimation for 3 point line
• Switch to Colour Segmentation for teams