I once used this technique to make a 2D pseudo-lidar sensor from a camera mounted on a indoor vehicle, it worked fairly well when used on uniformly textured flooring. After transforming the camera input image using an IPT, you could do some basic trig to figure out the distances away from the camera to non-floor objects.
The first few are simple, photograph + DEM (digital elevation model) or DTM (digital terrain model) aka height map will give you a 3D landscape with photo-accurate texture. They add manual CAD modeling in the later images, and that gives them the extra detail they need for 3D buildings / structures.
For the interior, it's a lot more impressive. They're doing photogrammetry or SLAM equivalent, reconstructing depth from stereo photographs and stitching multiple pairs together to generate a scene. They're also adding CAD in that case, as you can see with the wireframe for the generator.
The title is missing a right parenthesis after the date https://www.xkcd.com/859 . EDIT: I originally put one in this comment, but then realised that that would be even worse once the title was fixed ….
Great stuff. Would it kill this site to allow for even just the tiniest bit of context for why a 22 year old article is being posted? Also it looks like fmwconcepts.com has come up on Hacker News a few times for Fred's ImageMagick scripts. But has anyone found the code to do the perspective transformation discussed here?
