Autocalibration of Cameras with Known Pixel Shape

Abstract

We present new algorithms for the recovery of the Euclidean structure from a projective calibration of a set of cameras of known pixel shape but otherwise arbitrarily varying intrinsic and extrinsic parameters. The algorithms have a geometrical motivation based on the properties of the set of lines intersecting the absolute conic. The theoretical part of the paper contributes with theoretical results that establish the relationship between the geometrical object corresponding to this set of lines and other equivalent objects as the absolute quadric. Finally, the satisfactory performance of the techniques is demonstrated with synthetic and real data.