2019/08/22 by Dinh-Cuong Hoang, Hoang, Dinh-Cuong, Todor Stoyanov +3
Computer Science · Engineering · #Advanced Neural Network Applications #Advanced Vision and Imaging #FOS: Computer and information sciences #Robotics (cs.RO) #Robotics and Sensor-Based Localization
paper · pdf · doi:10.48550/arxiv.1908.08601
openalex publication_date 2019/08/22 · openalex created_date 2022/07/28 · openalex updated_date 2026/07/28
We present an approach for recognizing all objects in a scene and estimating\ntheir full pose from an accurate 3D instance-aware semantic reconstruction\nusing an RGB-D camera. Our framework couples convolutional neural networks\n(CNNs) and a state-of-the-art dense Simultaneous Localisation and Mapping\n(SLAM) system, ElasticFusion, to achieve both high-quality semantic\nreconstruction as well as robust 6D pose estimation for relevant objects. While\nthe main trend in CNN-based 6D pose estimation has been to infer object's\nposition and orientation from single views of the scene, our approach explores\nperforming pose estimation from multiple viewpoints, under the conjecture that\ncombining multiple predictions can improve the robustness of an object\ndetection system. The resulting system is capable of producing high-quality\nobject-aware semantic reconstructions of room-sized environments, as well as\naccurately detecting objects and their 6D poses. The developed method has been\nverified through experimental validation on the YCB-Video dataset and a newly\ncollected warehouse object dataset. Experimental results confirmed that the\nproposed system achieves improvements over state-of-the-art methods in terms of\nsurface reconstruction and object pose prediction. Our code and video are\navailable at https://sites.google.com/view/object-rpe.\n