Exploitation of GPS-control points in low-contrast IR-imagery for homography estimation

This paper gives a novel approach to homography estimation. It concentrates on the problem when there are no distinct features that can be matched between two consecutive images. The proposed method exploits knowledge of the camera system and sensors located on an aerial platform that is used to capture the infrared imagery. Exploiting additional information from a GPS receiver, an inertial measurement unit, and a compass, the position and direction of the camera can be determined. Once the information is available the projection mapping can be computed and real point targets can be inserted into the individual video frames. This technique yields specific point correspondences with which the homographies can be computed. The resulting homographies describe an improvement compared to homographies based solely on image features. The proposed method will be compared with sophisticated methods working with image features, e.g. gray values based methods. The aim of the method given in this paper is to overcome problems like drifting and the ambiguity of the pixel movements.