Are there any efforts to 3D-map cave systems? It seems like such a no-brainer application of depth-sensing technologies and yet it seems to me nobody is doing it! What gives?!
The lidar on the phones suck (I had one and was super disappointed). It has a range of like half a room, terrible resolution, was slow as heck, and the software was buggy and unstable and the algorithms often butchered all but the simplest scans. It's not like you put the phone on the tripod and you get a Ubisoft-like cave in a minute. It's like you get half the room at 20% fidelity after an hour of frustration.
Honestly the more mature photogrammetry techniques that stitch together multiple photos (from video or 360 shots taken successively) seemed much more impressive. Indeed, you can run most of the same apps with lidar off and still get similar results. There is a lot of depth info when you take pictures from different angles that a smart algorithm can figure out. The lidar adds more hinting but without professional grade equipment, the optical sensors (cameras) are much higher resolution and range.
The lidars on the i devices are cheap gimmicks.
Edit: just to add a bit more detail, lidar basically gets you a point cloud of depth data. The professional ones have big spinning mirrors that blanket the room in a matrix of powerful dots, like a Kinect on steroids. That takes a lot of power and, usually, moving parts.
The iPhone one uses more modern, smaller solid state technology but has a resolution of a few hundred dots and a range of 5m, vs the millions of pixels on the optical camera and a range of however much ambient light you have.
The lidar is an active emitter. It has to produce its own laser light, limited by its emitters and battery, and have a way to scan its emissions across 2d space (x and y axes) or else have a ton of micro lasers in a grid.
By comparison the optical camera is just a passive receiver, accepting reflected sunlight or light from the the weak LED flash. Those light sources are way stronger and more diffuse in terms of range and coverage.
Maybe in a few years the lidar will get better but for now it's just not useful for anything except face ID. Prosumer drones probably have better lidar and also better photogrammetry software.
You really think those are fantastic? The only ones that seem even remotely good are the knew where the camera rotates around a static object (couch, whatever). For the room scale ones, the results are terrible. Like this one of the sketch fab hq: https://sketchfab.com/3d-models/sketchfab-hq-scanned-with-ip...
It looks like something out of a video game dream sequence where your world is falling apart.
They are roughly comparable with results from photogrammetry at a fraction of the capture/processing time. Have you seen much raw photogrammetry output? Before it's been repaired/cleaned-up? You can get better than this with photogrammetry but it takes a lot of time and trial and error. This stuff is as good as my personal attempts at photogrammetry and in some cases better. When photogrammetry fails it generally fails catastrophically. With LIDAR you will always get something.
I think we generally agree. I'm just saying it's not good enough to take into a cave and casually scan it into a 3d model yet, not with the cheap consumer tech at least. I don't think those results are usable for anything (games, sharing, measurements, mapping), just novelty.
Actually I don't think we do agree. "Novelty" is a very low bar and I'm arguing that there's plenty of utility above and beyond that.
Measurement? Yes - it's accurate enough for many purposes like home furnishing and interior design mockups.
Games? Yes. It suffers from similar issues to photogrammetry (high poly count, need for repair and cleanup) but that's proved useful in many aspects of game content creation. There's also the fact that games are a creative medium and people develop new art styles all the time. There's some very creative uses of point clouds and 3d volumetric capture which is often glitchy as hell.
Sharing? Not sure what you mean.
Mapping? It's much better quality than the 3d meshes in Google Maps and they are incredibly useful for visualising and inspecting places you can't visit. Compare with the output from drone photogrammetry. This plugs a gap between large scale mapping via drones and laser site scans. It's better than the former and cheaper than the latter.
It's not going to replace industrial laser scanners for surveying and site inspection but it's good enough for many of things we're currently using photogrammetry for.
And the sheer ease of use could open up new applications that wouldn't have been viable with slower workflows. Speed is a feature in it's own right.
The standard cave survey technique is fiberglass tape measure, magnetic compass, and inclinometer, visually sighted point-to-point; passage width and height are recorded at each station in two planes, reflecting the traditional 2D cave map projections of plan and profile.
These days, the most common mapping technique is a similar workflow, but with a ruggedized handheld laser distance meter with digital compass and inclinometer. They may have on-device storage, and they may transmit measurements to a PDA or tablet via Bluetooth. Some people sketch the 2D projections of the cave on that digital lineplot, some use the data to sketch with pencil on paper. A cave is a very hostile environment to electronics, so "caveman technology" is often intentionally preferred.
Lidar and photogrammetry are quite difficult in a cave. The surfaces are irregular, of varying wet and muddy conditions whose reflectivity cause problems. And shadows - both visible and line-of-sight shadows - produce "swiss cheese" 3D models in all but the simplest caves. Lava tubes are often conducive to these techniques; their basalt composition wreaks havoc on magnetic compasses, so it's a welcome update.
Having used a $75k FARO Lidar device in an admittedly clean and comfy cave before, I sure wouldn't attempt to haul it into the average "wild" cave for fear of trashing it. The promise of SLAM-based devices, a computationally-expensive technique for realtime integration of 3D data building a 3D model as you traverse the cave, is amazing; but the expense still high, the hardware still fragile, and caves are still irregular enough to never make this easy.
Stone Aerospace in Texas has done some amazing work with autonomous underwater cave mapping using sonar. Their submersible robot has the same advantage that a quadcopter would have in an air-filled cave, whereas we human mappers are still stuck standing on (or crawling on) the floor.
