2020/08/06 by Jana Pavlasek, Pavlasek, Jana, Stanley M. Lewis +5 · 1 citation
Computer Science · Engineering · #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Hand Gesture Recognition Systems #Handwritten Text Recognition Techniques #Robot Manipulation and Learning #Robotics (cs.RO)
paper · pdf · doi:10.48550/arxiv.2008.02881
openalex publication_date 2020/08/06 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
Robots working in human environments must be able to perceive and act on\nchallenging objects with articulations, such as a pile of tools. Articulated\nobjects increase the dimensionality of the pose estimation problem, and partial\nobservations under clutter create additional challenges. To address this\nproblem, we present a generative-discriminative parts-based recognition and\nlocalization method for articulated objects in clutter. We formulate the\nproblem of articulated object pose estimation as a Markov Random Field (MRF).\nHidden nodes in this MRF express the pose of the object parts, and edges\nexpress the articulation constraints between parts. Localization is performed\nwithin the MRF using an efficient belief propagation method. The method is\ninformed by both part segmentation heatmaps over the observation, generated by\na neural network, and the articulation constraints between object parts. Our\ngenerative-discriminative approach allows the proposed method to function in\ncluttered environments by inferring the pose of occluded parts using hypotheses\nfrom the visible parts. We demonstrate the efficacy of our methods in a\ntabletop environment for recognizing and localizing hand tools in uncluttered\nand cluttered configurations.\n