This paper maps the Pamir shipwreck off Barbados over several dives with an affordable action camera, adding water depth from a dive computer to the visual-inertial data. SVIn2 produces a trajectory and sparse map per session, COLMAP optimises globally and builds a dense reconstruction, and calibration targets at fixed spots bring the sessions into one frame. The exterior and the accessible interior of the wreck were mapped in two sessions, and a third session used two cameras with different fields of view.
Relevance
Multi-session alignment under water still leans on fixed targets, which shows how hard revisits are without GPS. Read for underwater perception and long-term autonomy.
Links
- arXiv
- Subfield: Field robotics, Mapping and localisation