Key Points
We extract the ridgeline separating the ground and sky from 360-degree images and match it against ridgeline signals paired with known locations. This uses views of the terrain in all directions to estimate the current location, even in sand dunes with few distinctive surface textures.
Abstract
It is difficult to identify where a photograph was taken in sand dunes because there are few landmarks, such as buildings or distinctive surface patterns. We instead use the shape of the boundary between the dunes and the sky. This boundary is called the ridgeline.
A 360-degree camera captures the surroundings, allowing us to extract the ridgeline in every direction. We compare this shape with a collection of ridgeline records paired with known locations and find the closest match. The location attached to that record gives an estimate of the camera's current position.
We tested the method using both computer-generated images based on terrain-height data and real 360-degree images taken in sand dunes. In both tests, it estimated locations more accurately than the existing method used for comparison. The approach uses the surrounding terrain's outline as a cue even where distinctive surface patterns are scarce.
Publication
- Shogo Fukuda, Shintaro Nakatani, Masashi Nishiyama, Yoshio Iwai,
Geo-localization using Ridgeline Features Extracted from 360-degree Images of Sand Dunes, Proceedings of 15th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP), Vol. 5: VISAPP, pp. 621 - 627, February 2020.
