A Novel Approach for Stereo-Matching Based on Feature Correspondence

摘要

A dense two-frame stereo matching technique generally uses an image pair as input in addition to the knowledge of disparity range. For realtime computer vision systems, however, there is lots of information that can enhance stereo-correspondence, e.g. feature points. Feature correspondence is essential for computer vision applications that require structure and motion recovery. For these applications, disparity of reliable feature points can be used in stereo-matching to produce better disparity images. Our proposed approach deals with adding the feature correspondences effectively to dense two-frame stereo correspondence framework. The experimental results show that the proposed approach produces better result compared to that of the original algorithm RecursiveBF [1].