Low-altitude high-resolution images have great potentials in many applications. However, there still have many difficulties in low-altitude image matching, such as severe geometric deformation and repetitive structure, etc. In this paper, we present a novel matching approach within the constraints of affine invariant regions and geometric relationship to obtain a substantial number of true matches from urban low-altitude images taken from different viewpoints. The basic idea of our method is to match feature points by using the geometric relationships between corresponding pairs, which can be represented by the fundamental matrix. Therefore, the precision of estimated fundamental matrix plays a crucial role in the whole matching and the proposed method can be composed of two parts: 1) Estimation of the fundamental matrix based on affine-invariant regions. 2) Matching the features within the estimated epipolar line and the new strategy denoted closest-to-1. After the two stages, a substantial number of true matches can be achieved. Experimental results of two groups of typical urban low-altitude images show that the proposed method has a big advantage over the other methods in the number of correct matches and matching accuracy.