A novel non-central catadioptric auto-calibration approach

Calibration is the determination of coordinates of all pixels' rays in some common coordinate system. It enables to compute a 3D ray, along which light travels, for every image pixel. In this paper we propose a novel non-central catadioptric system auto-calibration approach. It uses polarization imaging to know the specular surface shape in order to estimate the catacaustics of the catadioptric system which geometrically fully describe the imaging system. The catacaustic is a locus of viewpoints. Each pixel in the image maps to a point on the caustic surface on which every point maps to a unique light ray from the scene. Polarization imaging helps to widen the range of catadioptric systems to include all types of specular surfaces. Our novel approach helps to auto-calibrate any combination of specular surfaces and lenses.