| Abstract |
CCTV camera calibration and geolocalization are essential for the deployment of smart-city applications. Despite the pressing need for practical and accurate surveillance camera calibration, most existing techniques are cumbersome and rely on restrictive assumptions on the camera’s or the scene’s geometry. In this context, we propose CCTV-Calib, a light and user-friendly toolbox to calibrate traffic cameras using satellite views. Unlike other surveillance camera calibration techniques, CCTV-Calib can estimate the intrinsic and extrinsic parameters as well as the GPS location of one or multiple CCTV cameras in a few clicks. For this purpose, we propose a simple yet effective two-stage pipeline to ensure a robust, repeatable, and easy calibration process. In the first stage, we perform automated keypoint matching between the CCTV image and the satellite views to compute an initial estimation of the camera’s parameters. In the second stage, this rough initialization guides a fine matching strategy, further improving calibration accuracy. This novel calibration pipeline is integrated into an easy-to-use GUI, making traffic camera calibration accessible to non-computer vision experts. To qualitatively and quantitatively evaluate the accuracy and relevance of our technique, we propose a novel dataset composed of synthetic and real images captured around the globe. Finally, in order to illustrate the pertinence of our calibration strategy, we demonstrate its applicability to 3D vehicle geolocalization. We made this toolbox and datasets freely available: https://github.com/rameau-fr/CCTV-Calib . © 2023, The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature. |
