Phase-Correlation Guided Search for Realtime Stereo Vision

In this paper; we propose a new theoretical framework, which is based on phase-correlation, for efficiently solving the correspondence problem. The proposed method allows area matching algorithms to perform at high frame rates, and can be applied to various problems in computer vision. In particular, we demonstrate the advantages of this method in the estimation of dense disparity maps in real time A fairly optimized version of the proposed algorithm, implemented on a dual-core PC architecture, is capable of running at 100 frames per second with an image size of 256 x 256.