duy-phamduc68/TrafficLab-3D

What it solves

TrafficLab 3D simplifies the creation of "digital twin" traffic simulations from standard CCTV footage and Google Maps data. It removes the need for professional camera calibration tools or high-quality synchronized satellite imagery, making advanced computer vision analysis accessible to students and independent researchers.

How it works

The system operates through a three-stage pipeline:

1. Calibration: It establishes a two-way projection between the CCTV view and a satellite map. This involves undistorting the lens, computing homography via RANSAC, and calculating camera position using parallax (head and ground contact points).
2. Inference: The tool runs object detection (e.g., YOLO) and tracking models on the footage to identify vehicles and pedestrians, saving the results as compressed JSON files.
3. Visualization: A rendering engine displays a synchronized side-by-side view, showing 3D bounding boxes on the CCTV feed and corresponding floor boxes, speed, and orientation on the satellite map.

Who it’s for

• Students and academic researchers studying traffic flow.
• Individual investigators and enthusiasts interested in computer vision.
• Developers building digital twin demonstrations for urban environments.

Highlights

• End-to-end workflow: Covers everything from initial camera calibration to final 3D visualization.
• Flexible Model Integration: Allows users to easily swap different object detection and tracking models.
• Accessible Calibration: Uses a stage-based process to map CCTV pixels to real-world satellite coordinates without specialized hardware.
• Digital Twin Experience: Provides synchronized views of the original footage and a 3D mapped representation.