3D semantic interpretation for robot perception inside office environments

• A robust registration framework for 3D point cloud data.
• Method for estimating the principal axes of 3D complete indoor scans.
• Improved hierarchical model fitting by clustering inliers.
• Labeling important indoor components, without the use of training classifiers.
• A technique for quadrilateral-like shape estimation.

Making sense out of human indoor environments is an essential feature for robots. The paper at hand presents a system for semantic interpretation of our surrounding indoor environments such as offices and kitchens. The perception and the interpretation of the measured data are essential tasks for any intelligent system. There are different techniques for processing 3D point clouds. The majority of them include acquisition, iterative registration, segmentation, or classification stages. We describe a generic pipeline for indoor data processing and semantic information extraction. The proposed pipeline is validated using several data sets collected using different 3D sensing devices.