Ground Penetrating Radar (GPR) is a promising sensor for landmine detection, but there are two major problems to overcome. One is the non-planer(e.g. rough and/or undulating) ground surface. It remains irremovable clutters on a sub-surface image output from GPR. Geography adaptive scanning is useful to image objects beneath non-planer ground surface. The other problem is the distance between the antennas of GPR. When imaging the small objects such as an anti-personnel landmine close to antennas, it increases the nonlinearity of the relationship between the time for propagation and the depth of a buried object. In this paper, we modify Kirchhoff migration so as to account for not only the variation of height and pose of the sensor head, but also the antennas alignment of the vector radar. The validity of this method is discussed through application to the signals acquired in experiments.