2018/12/10 by Karthik Desingh, Desingh, Karthik, Shiyang Lu +5
Computer Science · #Advanced Image and Video Retrieval Techniques #FOS: Computer and information sciences #Human Pose and Action Recognition #Multimodal Machine Learning Applications #Robotics (cs.RO)
paper · pdf · doi:10.48550/arxiv.1812.03647
openalex publication_date 2018/12/10 · openalex created_date 2022/08/01 · openalex updated_date 2026/07/28
Robots working in human environments often encounter a wide range of\narticulated objects, such as tools, cabinets, and other jointed objects. Such\narticulated objects can take an infinite number of possible poses, as a point\nin a potentially high-dimensional continuous space. A robot must perceive this\ncontinuous pose to manipulate the object to a desired pose. This problem of\nperception and manipulation of articulated objects remains a challenge due to\nits high dimensionality and multi-modal uncertainty. In this paper, we propose\na factored approach to estimate the poses of articulated objects using an\nefficient nonparametric belief propagation algorithm. We consider inputs as\ngeometrical models with articulation constraints, and observed RGBD sensor\ndata. The proposed framework produces object-part pose beliefs iteratively. The\nproblem is formulated as a pairwise Markov Random Field (MRF) where each hidden\nnode (continuous pose variable) is an observed object-part's pose and the edges\ndenote the articulation constraints between the parts. We propose articulated\npose estimation by Pull Message Passing algorithm for Nonparametric Belief\nPropagation (PMPNBP) and evaluate its convergence properties over scenes with\narticulated objects.\n