GroupInteractionTracker

GroupInteractionTracker is a Python project for real-time person detection, pose estimation, and group interaction visualization from an RTSP camera feed.

Features

• Person detection and tracking using YOLO and ByteTrack.
• Human pose estimation and orientation calculation.
• Group interaction detection based on proximity and movement.
• Segmentation masks using SAM2 for more accurate person centroids.
• Overlay visualization with bounding boxes, IDs, orientation, and group lines.

Requirements

• Python 3.10+ (Python 3.11 recommended)
• GPU recommended for best performance with YOLO and SAM2
• models/yolov8n.pt model file present in the repository
• Dependencies installed from requirements.txt

Setup

1. Open a terminal in the repository root:

cd D:\texa\GroupInteractionTracker

2. Create and activate a virtual environment:

python -m venv venv
.\venv\Scripts\Activate.ps1

3. Install the required packages:

pip install -r requirements.txt

Configuration

Update src/config.py with your RTSP camera connection and settings:

RTSP_URL = "rtsp://username:password@camera_ip"

You can also adjust:

• FRAME_WIDTH and FRAME_HEIGHT
• CONFIDENCE
• SAM_UPDATE_INTERVAL
• GROUP_DISTANCE
• INTERACTION_TIME

Run

From the repository root, run:

python src/main.py

The application reads frames from the RTSP stream, performs detection, pose estimation, mask extraction, and displays results with group interaction overlays.

Project Structure

• src/ - Main Python source code
• models/ - YOLO model files
• assets/ - Additional resources used by the project
• logs/ - Output logs
• requirements.txt - Python dependencies

Notes

• The project uses ultralytics for YOLO tracking and the SAM2 repository for image segmentation.
• Make sure the model files are available and the RTSP stream is accessible before running.
• For GitHub, commit the README and any project changes together.