On Texture and Geometric Feature Based RGB-D Registration

Kurzfassung

This work addresses the problem of point cloud registration in the context of RGB-Depth data. The Color and Geometric Feature (CGF) registration presented here utilizes both intensity and depth data, in order to achieve an optimal use of all available resources. Registration is achieved by performing optimization over a joint cost function, exploiting texture features and correspondences between planes. For this, a group of planar patches is extracted and matched, with covariances estimated based on the sensor’s error model. Comparisons to other approaches highlight the advantages of CGF registration.