Context and feature sensitive re-sampling from discrete surface measurements

Abstract

This paper concerns context and feature-sensitivere-sampling of workspace surfaces represented by 3D pointclouds. We interpret a point cloud as the outcome of repetitiveand non-uniform sampling of the surfaces in the workspace. Thenature of this sampling may not be ideal for all applications,representations and downstream processing. For example it mightbe preferable to have a high point density around sharp edges ornear marked changes in texture. Additionally such preferencesmight be dependent on the semantic classification of the surfacein question. This paper addresses this issue and provides aframework which given a raw point cloud as input, produces anew point cloud by re-sampling from the underlying workspacesurfaces. Moreover it does this in a manner which can be biasedby local low-level geometric or appearance properties and higherlevel (semantic) classification of the surface. We are in no wayprescriptive about what justifies a biasing in the re-samplingscheme - this is left up to the user who may encapsulate whatconstitutes "interesting" into one or more "policies" which areused to modulate the default re-sampling behavior.