A special software of monocular depth estimation fashions
A number of approaches to estimating floor elevation from a single picture have been mentioned within the literature. In a earlier article, I mentioned how it’s doable to foretell the depth of a single 2D picture utilizing a monocular estimation mannequin. Nevertheless, when the enter to the mannequin is a picture of a specific floor, the prediction represents a Digital Elevation Mannequin (DEM). In my first analysis paper I confirmed how a DEM of the floor of Mars could be obtained from 2D greyscale pictures utilizing deep studying approaches. To raised perceive the concept I’m going to suggest, I recommend you first to strive the demo of the venture here.
Introduction
As mentioned in additional element in one other story, the DEM of a floor is a grid of elevation values the place every cell shops the elevation of a specific level on the floor:
DEMs are often represented graphically utilizing colour maps. Within the picture above, the very best factors are pink and the bottom factors are purple.
However, monocular depth estimation fashions are used to estimate the gap of every pixel of a picture from the digicam (even the digicam of a satellite tv for pc) that took the picture:
The thought is {that a} satellite tv for pc picture of a floor could be fed right into a monocular depth estimation mannequin. On this method, it’s doable to foretell the DEM of that floor, as a result of every level of the output represents a distance (depth), and elevations of the floor could be derived by utilizing depths (extra on this later).
The tactic mentioned on this article can be utilized for different surfaces as nicely, and never only for Mars.
UAHiRISE
The Excessive-Decision Imaging Science Experiment (HiRISE) is a digicam on board the Mars Reconnaissance Orbiter (a spacecraft that gives assist for…
