Motion estimation based on iterative color matching and structure matching

Motion estimation in detail is required by many vision applications. Conventional method such as optical flow method is often used to calculate the motion between two image frames. Color based constraint such as the assumption of brightness constancy underlies most optical flow estimation methods. However, it is impossible to calculate precise optical flows from only color value, since changing view point and inconstant illumination make the colors of some regions of objects entirely different between consecutive frames. The structure based constraint of object is very necessary for tracking as well. In our proposed method, both the color matching and structure matching are formulated by linear assignment problem and motions are obtained by solving them alternately.