Improved feature matching technique in low overlap area

Low overlap area is one of the problems in finding matching between two frames. When the camera moves on a static planar, the structure of the image deteriorates. As the camera moves, the same object might appear different with different camera views. Thus, it causes matching failure in subsequent images. We propose a rotation translation approach to assist existing feature matching algorithm. The method consists in rotating the image through a rotation translation model and applying Singular Value Decomposition (SVD) scheme and histogram patch similarity for filtering. The experimental results show that adding the proposed algorithm improves considerably the results in case of very small overlap region.