MSc Thesis Defense by Yibo Dang


Image analysis of farmland field based on homography


In this thesis, a method of global homography estimation based on a series images is discussed and proposed. The method aims at doing global image registration for the image series of a huge field and establishing a global coordinate system of that area. The target field in the project is the cereal farmland in Denmark. During the initial stage, feature detection and matching are discussed in order to catch the basic information for local homography estimation. Then, a linear equation system, which is integrated with the local homographies, is solved. The solution combines with several homography matrices. These homographies describe the relations between point locations in the single image frames and the corresponding coordinate in the global system. An endurable error rate for further applications, e.g. global thistle detection in farm field, is indicated from the experiments.

Supervisor: Søren Olsen

Censor: Søren Overgaard