A Survey of Planar Homography Estimation Techniques

The estimation of the homography between two views is a key step in many applications involving multiple view geometry. The homography exists between two views between projections of points on a 3D plane. A homography exists also between projections of all points if the cameras have purely rotational motion. A number of algorithms have been proposed for the estimation of the homography relation between two images of a planar scene. They use features or primitives ranging from simple points to a complex ones like non-parametric curves. Different algorithms make different assumptions on the imaging setup and what is known about them. This article surveys several homography estimation techniques from the literature. The essential theory behind each method is presented briefly and compared with the others. Experiments aimed at providing a representative analysis and comparison of the methods discussed are also presented in the paper.