Mapping step-by-step in DefSLAM
Source:
lamarca-2020
Parent:
DefSLAM framework
Steps
- 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
- [
NRSfM
] Process k* to get estimate of an up-to-scale surface
- Input of NRSfM: warps
- [ 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
- [ SfT template substitution ] A new template (mesh with embedded map points) is created from the surface