A Simple Algorithm for Object Location from a Single Image without Camera Calibration

摘要

In this paper we propose a method for locating a point (or object) in space without any camera calibration or object model. Instead we assume that the 2D location of 4 coplanar points is known and that the unknown point lay on this plane. This is similar to computing the fundamental matrix linking pixel position in the image plane with another plane position in 3D space. A concrete example could be the accurate location of a car on a highway from an airplane or helicopter camera knowing the location of four milestones or the location of a man in a city from a surveillance camera knowing the location of four block corners. Other applications in domains such as photogrammetry, remote sensing or military reconnaissance are also possible just to name a few. The algorithm is illustrated and tested on real images and shows errors less than 2%.