Member-only story
How to Generate 3D models from 2D images?
First things first, let's start with some definition and concept explanation.
What Is Photogrammetry?
Despite its fancy name, photogrammetry is just the collection of precise information about structures or environments based on numerous image recordings.
Aerial photogrammetry and terrestrial photogrammetry are the two forms of photogrammetry. Aerial scanning occurs when the area is scanned from a device in the air (e.g. using a drone), and terrestrial scanning occurs when the area is scanned from the ground (e.g. using a tripod).
How Does It Work?
How can an algorithm create 3D models from just photos? The quick answer is that photogrammetry uses 3D geometry information and formulas e.g. the camera positions, its intrinsic parameters, and rays intersections …
To explain this further, if we take a photo from a dataset, every pixel in that image represents a light ray in 3D space that extends from the camera center to the real 3D point. In the figure below the three traced rays are from three different images that correspond to the same 3D point in the world.
