CSIRO/Data61 have demonstrated with their Zebedee system [1], that the precision mechanical system is optional. Their handheld LIDAR head bounces around randomly and the system is able fit and register the new measurements within the 3D model that it is progressively building.
Such a handheld LIDAR would be a very cool hobby/amateur project, as the mechanical side can be simple and low cost, making the capital cost low. The smarts are in the software and maths. There is also that fascination/aha factor of working though the maths and realising that smart processing means that some "obvious" things, such as precision mechanics, aren't actually necessary, then seeing it work.
At GeoSLAM (joint venture with CSIRO) we sell scanners and software based on the Zebedee (ZEB1) and a mechanically rotating version (the ZEB-REVO [1]).
The rotating version is easier to use (don't have to worry about nodding the head too fast or too little) while still having the same cost/speed advantages over terrestrial scanners.
Great project! but the title "3D scanning like a pro" is quite a stretch. Real 3D scanning pros use something like this Leica P40 scanner:
https://www.youtube.com/watch?v=GnmerVV5-T4
The definition of pro varies! Here's a short rundown of 3D imaging:
Laser triangulation systems (laser stripe + turntable) will give you extremely highly detailed models at close range. You could build one for under £100 and the accuracy can easily be sub-mm. Downside is it'll only work for a short range.
Long range you're not going to beat a scanning LIDAR system like the sort that Leica offers. A comparable system from Faro (with a public price-tag) is around £30-50k depending on spec. For 'set and forget' scanning, these systems are great. They're also pretty much the only reliable solution for robust, kinda fast capture of large sites. Velodyne's automotive LIDAR (the one on the older Google cars) came on the market for around $75k. The future of 3D imaging is probably low cost MEMS scanning LIDAR, which is being bankrolled by autonomous driving.
Short range you also have time of flight (Kinect 2) which is used in research and also by several large companies for commercial RnD (e.g. Ocado - http://robohub.org/wp-content/uploads/2017/02/SoMa-Ocado-RBO... - see the thing on the tripod?). My local second hand stores sells the Kinect 2 for about £25, but you need the USB adaptor (£30). Lots of companies make ToF sensors, but Microsoft really forced the price down.
Then you've got Kinect 1-like systems (Primesense), which include Google Tango (and other clones from Asus, etc). I think the Intel Realsense does something similar. These are mono cameras and an IR pattern projector. They don't work in sunlight, but are pretty good indoors. The Kinect 1 was (is) huge in 3D vision research because it was so cheap and relatively robust.
Finally you've got stereo which is the only real competitor to LIDAR outdoors. Much more dense than LIDAR, and you get true intensity information, but poor accuracy at long range. Performance is mostly dependent on the distance between the cameras. It's also the only technology you can reliably use in space. There are very few spaceborne LIDAR systems (e.g. LOLA, MOLA, etc). Stereo is the bread and butter of 3D on rovers and for planetary mapping. Prices vary, but you're looking at around £400-500 for a good pre-built stereo rig.
I think it would be unfair to say NASA and ESA (Exomars should also have stereo cams) aren't pros!
There are also several niche industries like handheld structured light scanners which can also get sub-mm accuracy. These are used for a mix of industrial inspection, capturing props for movies, etc.
You can also take the Kinect 2 apart and modify it for using external power input and regular USB3 (Type B?) cable.
I have done that and it works well, see: http://www.tatsch-it.de/kinect-2-usb-3-modification/
Mikeha, glad to see attention from my favorite 3D scanning company. Results you shared are awesome!
I do admit that my system lacks precision, angular resolution, and lot's of other nice bells and whistles. But hey it was build over weekend including software. Also this is more like rotary stage demonstration rather full featured 3D scanner.
thanks, that does put the OP's work, impressive as it is, into perspective, it's a stunning system you have there and looks like you get to work in some nice venues :)
And since we're in shameless plug territory already:
By comparison DotProduct [1] is doing full 3D model creation on tablets since 2013 (our first product was based on a Google Nexus 7 with just 1 GB of RAM -- no cloud required).
Maritime Archaeologist here: we take photos with a DSLR camera of terrestrial and underwater sites, and artifacts, and then extrapolate 3D models with Agisoft Photoscan.
Is there any advantage to using one of these devices?
> Is there any advantage to using one of these devices?
Yes, you can get the fully processed 3D model within minutes of capture, no cloud or desktop computer required. Plus our file format is very compressed so you can share scans right from the field over your mobile connection.
Another way of doing this is something like Google's project tango, right? [0][1]
So you can build and code this up (which sounds like a fun project) or if you're lazy or lack the time you can buy a $512 portable device that probably has an app or two to do 3D scanning out of the box.
Both approaches have their pros/cons but both are pretty neat and it's exciting that it should only be cheaper and easier for hobbyists to play with in the years to come.
Some of the visual SLAM techniques are also getting fairly good results for cheap. In particular, see DSO (same research group as the previous LSD-SLAM).
Is there any economical (we'll say sub-$2K) method that an individual can produce 3D scans of typical interior residential spaces? I've taken a lot of 2D photos of various places that are sentimental to me, but I'd love to make even low-quality 3D scans if an economical option existed.
The Matterport product mentioned elsewhere in the comments seems interesting, but they specifically disclaim that they don't offer a local hosting option. Since this is for personal use I really just want the data from the scans, not a recurring service model.
Edit: I'll clarify. Is there a "packaged" scanning hardware product in the sub-$2K range? I don't follow this technology closely and I'm really just sticking my head in to see where things are right now? I'm all for cobbling together software to deal with the resulting output, but I'm not in a position where I can cobble together hardware.
I don't know enough of the associated terminology, so I sound wholly uninformed here. What I see in the linked article are point clouds, which are definitely interesting, but I'm wondering if something exists to allow an individual to capture textured scans and store/view them on their own hardware. The Matterport product looks like what I'd want (at a $3.6K price-point for the camera, and arguably with higher quality output than I desire) but it sounds like it's really a service offering and doesn't just give you the resulting data for your own use w/o their recurring service.
Is this article not literally about doing 3D scans of interior spaces using cheap components? The stuff listed in the article costs $549 (plus a laptop/arduino/tripod etc).
I dropped on an edit. I'm a rank amateur here, and admittedly trying to use HN readers' knowledge as a substitute for research I should be doing on my own.
I built essentially the same kind of contraption for my PhD, except I used a gimbal scanner (±90 degrees in two axes) and an industrial 1D LIDAR (a Dimetix FLS-C). You get some interesting calibration problems, because if your LIDAR isn't perfectly centred on the rotation centre of both axes, you'll get offsets; in one direction you'll overshoot, in the other you'll undershoot. The basic trig works well enough for demonstrations though.
Similar downside: the LIDAR I used was extremely accurate (±1 mm stdev), but could only capture at around 20Hz. 500Hz would have been a luxury! I believe I could have pushed it to 200Hz with analogue out, but I wanted the accuracy.
had an idea to scan semi convex objects using a web cam and a green screen (something like this has probably been done before Im guessing). Just trace the outline of the object at various rotations (i.e. using a turn-table) and produce a model from the result. Probably do two passes, rotating the object 90 degrees on its side for the second pass.
There was a software in the 90s or early 2000s called "Sculptor 3D" that did precisely this.
You would print a sheet of paper that had location marks, then put in the center whatever object you wanted to model and take pictures of it from different angles.
Then you would manually select each marker (in order, so the program could determine where was this picture taken from) and finally you would outline the object you were interested in.
The program would then process everything and extract a 3D model out of the pictures including textures.
It was pretty amazing for the time, so I'm guessing nowadays this should be much more doable.
They even had a free version distributed in a 3D design magazine, unfortunately due to the very common name it's being very hard for me to find any references to it.
See also: Visual Structure from Motion, which I believe works off static light sources (same stuff you see in 123D Catch and similar software, this one is free though).
Bravo! This is really impressive! I have always wanted to do something like this with a kinect, but I think I like this method way more because of the lack of any black boxes. You can really see how the whole thing works, top to bottom.
Such a handheld LIDAR would be a very cool hobby/amateur project, as the mechanical side can be simple and low cost, making the capital cost low. The smarts are in the software and maths. There is also that fascination/aha factor of working though the maths and realising that smart processing means that some "obvious" things, such as precision mechanics, aren't actually necessary, then seeing it work.
[1] https://www.data61.csiro.au/en/Our-expertise/Expertise-Cyber...