This is a question recently raised by a friend of our knowledge planet "3D Vision from Beginner to Mastery": "Is there a more mature way to relocate on a 2D map (there are walls, doors and other information on the map) within a limited range through vision?"
In the answer, the guests of our planet, as well as the bigwigs, gave some of their own thoughts.
Then, in conjunction with a series of work that I've seen before, I'd like to expand on that a little bit more.
This is an article that has been studied before, and it is also a series of works. That is, in an unvisited environment, positioning is carried out through the room layout plan.
论文：LaLaLoc: Latent Layout Localisation in Dynamic, Unvisited Environments（ICCV 2021）
LaLaLoc++: Global Floor Plan Comprehension for Layout Localisation in Unvisited Environments (ECCV 2022)
Project address: https://github.com/ActiveVisionLab/LaLaLoc
The first is the first LalaLoc, which is positioned by an implicit representation of the room layout by learning a hidden space shared between the RGB panoramic image and the layout inferred from the known floor plan through the network, introducing direct cross-modal pose optimization in the hidden space.
Then there is LalaLoc++, which is an improved version of LaLaLoc, which requires the first use of a plane to construct an explicit 3D representation of the scene, a process that requires assumptions about the scene geometry, and LaLaLoc++ introduces a global plan understanding module that can intensively infer the scene structure directly from the 2D plan, eliminating the need for explicit modeling or rendering.
Interested readers can download two articles to study,They provide a way to locate through the room layout plan in an unvisited environment,Through a vision-only solution,The cost is lower,And the room layout plan is also a map form that occupies a small amount of storage,Its solution is very suitable for deployment on the sweeping robot。
Finally, the above discussion and related article sharing and interpretation can be found on our planet!