A view based approach for matching the 3D appearance of local features

Abstract

This paper presents a novel 3D RGB-D keypoint extraction approach for creating stable point-to-point correspondences between 3D free form objects under changes in view point, termed view compartmentalised keypoints or VC keypoints. VC Keypoints are extracted and matched based on the local directionally dependant appearances of the keypoints from a gamut of observed 3D poses, creating a 3D structure of keypoint appearance across the view space. In addition, we show how keypoints observed across a range of views can be integrated to form more comprehensive keypoint descriptors. The VC keypoint methodology shows an improvement in point-to-point matching over SIFT, SURF and EC-SIFT. © 2013 Springer-Verlag.