Abstract/Summary

We address the self-calibration of a smooth generic central camera from only two dense rotational flows produced by rotations of the camera about two unknown linearly independent axes passing through the camera centre. We give a closed-form theoretical solution to this problem, and we prove that it can be solved exactly up to a linear orthogonal transformation ambiguity. Using the theoretical results, we propose an algorithm for the self-calibration of a generic central camera from two rotational flows. In order to solve the self-calibration problem using real images, we also study the computation of dense optical flows from image sequences acquired by the rotation of a smooth generic central camera. We propose a method for the computation of dense smooth generic flows from rotational camera motions using splines. The proposed methods are validated using both simulated and real image sequences.