Calibration Methodology for Distant Surveillance Cameras

We present a practical method for video surveillance networks to calibrate their cameras which have mostly non-overlapping field of views and might be tens of meters apart. The calibration or estimating the camera pose, focal length and radial distortion is an essential requirement in video surveillance systems for any further automated tasks like person tracking or flow monitoring. The proposed methodology casts the calibration as a localization problem of an image with respect to a 3D model which is built a priori with a moving camera. The method comprises state-of-the-art functioning blocks, the Structure from Motion (SfM) and minimal Perspective-n-Point (PnP) solvers, which were proved stable in 3D computer vision community and applies them in context of video surveillance. We demonstrate that the calibration method is effective in difficult repetitive, reflective and texture less large indoor environments like an airport.