Mapping step-by-step in DefSLAM

Search IconIcon to open search

Source: lamarca-2020
Parent: DefSLAM framework

unknown_filename.1.png

Steps

  1. Recover warps between k and k* (s.  [Non-Rigid Guided Matching (b/w KFs) in DefSLAM](non-rigid guided-matching-(b_w-kfs)-in-defslam.md))
    • with k: anchor keyframes, i.e. KFs where one of the observed map points was initialised
    • with k*: set of best covisible keyframes
    • warps: transformation between the images Ik to Ik*
      • In DefSLAM, Schwarps (a family of warps using 2D Schwarzian equation regularisers) is used
      • Schwarps has something to do with the infinitesimal planarity assumption of NRSfM
  2. [ NRSfM ] Process k* to get estimate of an up-to-scale surface  unknown_filename.2.png
    • Input of NRSfM: warps
  3. [ Surface alignment ] Up-to-scale surface (\hat{S}_k) is aligned with the whole map in order to obtained the scaled surface Sk w.r.t. the old map in keyframe k
  4. [ SfT template substitution ] A new template (mesh with embedded map points) is created from the surface

unknown_filename.png