Robust Planar Pose Estimation by Viewpoint Generative Learning(<Special issue>Real World Imaging)

Abstract

We propose a planar pose estimation method that is robust to viewpoint changes. Conventional local features such as SIFT, SURF, etc., have scale and rotation invariance but often fail in keypoint matching when the camera pose significantly changes. To solve this problem we adopt viewpoint generative learning. By generating various patterns as seen from different viewpoints and collecting local features, our system can learn a set of descriptors under various camera poses for each keypoints before actual matching. Experimental results comparing usual local feature matching or patch classification method show both robustness and fastness of learning. Proposed method can achieve a markerless AR system that sets a tracking target on site.