We present a method that removes point cloud artifacts like noisy points, missing data and outliers from a point cloud using a learned shape prior. The shape prior is learned with the Gaussian Process Latent Variable Model from a set of reference objects. As input data our method uses the estimated object pose from an object detector and a segmented point cloud. We show that the estimated shape prior is capable of modeling fine details to a certain degree. We also show that after applying our method the measured accuracy and completeness is increasing.
展开▼
