Computing camera orientation relative to a world coordinate frame by detecting its projected axes

In this paper a new approach is presented to compute the camera orientation relative to some man-made objects coordinate frame, the world coordinate frame. The approach is based on the observation, that man-made structures expose many lines, which are aligned with three principal orthogonal directions, belonging to a cartesian world coordinate frame. Modeling the camera as an orthographic one, a nonlinear equation system is derived, which solutions yield three angles, that describe the camera orientation relative to the scene. In order to detect the projected principal axes of the world coordinate frame, an analysis of the log-magnitude spectrum of an image is used, followed by an algorithm that uses line information from the input image directly. The algorithm has been tested on a variety of rendered test images and real images of ships.