Our paper "Accurate Automatic Camera Calibration on Low-Quality CCTV Traffic Video Streams" has been published at the International Conference on Computers and Automation (CompAuto 2022).
The Challenge
Extracting vehicle trajectories from traffic cameras is valuable for urban planning and traffic analysis. However, most CCTV cameras are uncalibrated and produce low-quality video, making accurate trajectory extraction difficult.
Our Approach
We analyze video streams from 511 traffic surveillance cameras in varying areas and arbitrary relative locations and orientations to roads. We propose a novel automatic calibration methodology that transforms video streams to a top-down perspective from just a minute of video.
Key Contributions
- Fully automatic camera calibration without manual intervention
- Works with low-quality CCTV footage
- Requires only one minute of video for calibration
- Tested across 511 diverse camera setups
This is joint work with Audhav Durai and Sankarshanaram Vempati.