Homography estimation for low-contrast IR image sequences utilizing GPS control points

Abstract

Abstract Estimating the homographies for IR sequences of approaching objects in the infrared spectrum is a special problem. Due to the approaching movement of a helicopter towards measurement items, the objects are represented with few pixels in the beginning of the sequence and they are represented with a lot more pixels when reaching the destination area. This fact results in an assignment problem of finding point correspondences. This difficulty has led to drifting phenomena when using sophisticated methods for estimating the homographies. In this article a method is described which eliminates both the problem of finding point correspondences as well as the drifting issue. The presented procedure exploits information gathered by additional sensors (i. e., a GPS receiver, an inertial measurement unit (IMU), and a compass) for the computation of camera projections for each frame. The camera mappings are used to project GPS control points into the frames. The use of the same GPS points for each frame results in naturally given point correspondences. Using these correspondences the homographies can be estimated. The results are compared to the state of the art.