eduzhai > Applied Sciences > Engineering >

Unsupervised Learning of 3D Point Set Registration

  • Save

... pages left unread,continue reading

Document pages: 10 pages

Abstract: Point cloud registration is the process of aligning a pair of point sets viasearching for a geometric transformation. Recent works leverage the power ofdeep learning for registering a pair of point sets. However, unfortunately,deep learning models often require a large number of ground truth labels fortraining. Moreover, for a pair of source and target point sets, existing deeplearning mechanisms require explicitly designed encoders to extract both deepspatial features from unstructured point clouds and their spatial correlationrepresentation, which is further fed to a decoder to regress the desiredgeometric transformation for point set alignment. To further enhance deeplearning models for point set registration, this paper proposes Deep-3DAligner,a novel unsupervised registration framework based on a newly introduced deepSpatial Correlation Representation (SCR) feature. The SCR feature describes thegeometric essence of the spatial correlation between source and target pointsets in an encoding-free manner. More specifically, our method starts withoptimizing a randomly initialized latent SCR feature, which is then decoded toa geometric transformation (i.e., rotation and translation) to align source andtarget point sets. Our Deep-3DAligner jointly updates the SCR feature andweights of the transformation decoder towards the minimization of anunsupervised alignment loss. We conducted experiments on the ModelNet40datasets to validate the performance of our unsupervised Deep-3DAligner forpoint set registration. The results demonstrated that, even without groundtruth and any assumption of a direct correspondence between source and targetpoint sets for training, our proposed approach achieved comparative performancecompared to most recent supervised state-of-the-art approaches.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...