A Camera Recalibration Method for a Top-View Surveillance System based on Relative Camera Pose and Structural Similarity

In this paper, we present a camera recalibration method for a discontinuous top-view stitched image. A top-view stitched image using multiple camera images is efficient for surveillance camera monitoring because it enables monitors to understand surroundings easily by spatial continuity. However, the top-view stitched image loses the advantage even if only one camera used in the stitched image is shifted by physical contact or its own weight. A simple solution such as returning shifted cameras physically or calibrating cameras with calibration markers again takes time and personnel. To address this problem, we propose the recalibration method based on relative pose of camera and image structural similarity in the top-view image. The proposed method performs recalibration without changing camera pose physically or using calibration markers. As a result of an experiment, we found out that our method restores a discontinuous top-view image stitched closing to the original one.