Realtime stereo matching using maximally stable extremal regions

In this paper we present a novel stereo matching algorithm Cyclops. The algorithm uses Maximally stable extremal regions (MSER) for localizing parts of the image in rectified stereo image pair. Each detected MSER is normalized to a fixed size image. The normalized MSER regions are matched in the stereo image pair taking into account the epipolar line constraints. Matched MSER pairs are used to calculate the disparity image using the semi-global matching technique. Cyclops algorithm reduces stereo matching computation time and memory usage. The proposed algorithm is therefore useful for real time stereo vision.