Developing robot perception systems for recognizing objects in the real-world requires computer vision algorithms to be carefully scrutinized with respect to the expected operating domain. This demands large quantities of ground truth data to rigorously evaluate the performance of algorithms. This paper presents the EasyLabel tool for easily acquiring high quality ground truth annotation of objects at the pixellevel in densely cluttered scenes. In a semi-automatic process, complex scenes are incrementally built and EasyLabel exploits depth change to extract precise object masks at each step. We use this tool to generate the Object Cluttered Indoor Dataset (OCID) that captures diverse settings of objects, background, context, sensor to scene distance, viewpoint angle and lighting conditions. OCID is used to perform a systematic comparison of existing object segmentation methods. The baseline comparison supports the need for pixel-and object-wise annotation to progress robot vision towards realistic applications. This insight reveals the usefulness of EasyLabel and OCID to better understand the challenges that robots face in the real-world.

No abstract

International audienceIntelligent Autonomous Robots deployed in human environments must have understanding of the wide range of possible semantic identities associated with the spaces they inhabit – kitchens, living rooms, bathrooms, offices, garages, etc. We believe robots should learn this information through their own exploration and situated perception in order to uncover and exploit structure in their environments – structure that may not be apparent to human engineers, or that may emerge over time during a deployment. In this work, we combine semantic web-mining and situated robot perception to develop a system capable of assigning semantic categories to regions of space. This is accomplished by looking at web-mined relationships between room categories and objects identified by a Convolutional Neural Network trained on 1000 categories. Evaluated on real-world data, we show that our system exhibits several conceptual and technical advantages over similar systems, and uncovers semantic structure in the environment overlooked by ground-truth annotators

No abstract

No abstract