Perspective correction for improved visual registration using natural features.

This research proposes a perspective invariant registration algorithm which improves on popular registration algorithms such as SIFT and SURF by correcting for perspective distortion using optical flow. A novel addition to the natural feature based registration process is proposed, which uses orientation information from previously correctly registered frames to attempt perspective correction. This process is evaluated when applied to the Natural Feature algorithms described. This research overcomes the cause of registration failings based on perspective distortion in Natural Feature Tracking, and attempts to find a better resolution than just pruning invalid matches. The results show that the proposed algorithm improved registration on two prominent Natural Feature based Registration algorithms, SIFT and SURF.