Reliable image matching method based on self-adaptive triangle constraint

The determination of conjugate points in a digital image pair, referred to as 'image matching', is one of the critical factors limiting the complete automation of photogrammetry as well as the computer vision. Traditional matching methods continue to encounter obstacles in areas of forestation, settlements, water, and in regions with poor texture. However, during the relative orientation and automatic aerial triangulation, a series of well-defined points with good texture and proper distribution has, in fact, been produced. This study aims to utilize the well-de fined points information as prior knowledge to improve the image matching. Since the terrain surface is either continuous as a smooth surface, or discontinuous as an irregular or rough surface. We observe that most of the well-defined points are found on the edge of irregular areas. Thus, a new image matching method is proposed: (1) to triangulate the well-defined points based on the Delaunay triangle criterion and then to consider each triangle as a local continuous area, which is then used as a geometric constraint for image matching. (2) to integrate the triangle constraint with the conventional epipolar geometric constraint, and to process the matching propagation which is controlled by 'the best possible, the first matching' strategy. (3) The local continuity constraint is enforced by updating the constraint triangulation adapt to the local texture feature. Intensive experimental results prove that the reliability and the final accuracy of image matching are distinctively improved, compared to the current image matching method of digital photogrammetric workstation.