Anybody looked at ShapeScale? Full body scanner for $350 (CES 2018)

Just seen a video about CES 2018 and this product caught my eye.

https://shapescale.com/tech.html

It is a scanner that is for the fitness industry but it takes a full body scan of you (including textures). This could potentially make creating specific avatars for games much quicker as all the clothing is scanned too. And at $350 that is vastly cheaper than getting a character modeller to create a game ready model. Granted this would be high res mesh but it should be trivial to create a low poly version from it, rig it and off you go.

If any of you get to CES please check this out.

Thanks

I looked it up and the real quality doesn’t seem that impressive. https://www.engadget.com/2017/05/10/shapescale-3d-body-scanner/

Because of how the scanning works, the resolution can’t be that high because the person is moving while the scanning is happening. This is why you need an expensive camera rig setup that fires simultaneously currently for good full body scans.

If I was making a video game in 2001, maybe. :wink:

You can buy really cheap webcams (5 bucks) these days from China, that have a decent resolution. If you’re able to manually control and lock down their camera settings, you can probably do a basic rig that easily beats this in both quality and price.

You can have a fully programmable camera for about 25 euros. https://www.youtube.com/watch?v=o2NNepaVqXA (Raspberry Zero + Sony IMX219)

1 Like

Based upon the price and the fact that it includes a “robotic” (ahum), a rotating camera around the object.
This probably is not based upon a TOF camera (although they seam to use 3 lenses like kinect/asus camera’s.)
It’s likely they do advanced normal camera stitching since the way the camera moves that would be enough inputs to reconstruct a 3D image, (cheap DIY 3d camera object scanners use the same trick)…
It beats me why they seam to use a camera width 3 lenses though, as rotating camera’s don’t require anything special.
The only speciality would be the using of (known) Math.

1 Like

I tend to agree with ambi. To me this looks like pointless waste of money. Not impressive at all. It cannot work well no matter what hardware or math they use if the subject is moving while the scan is in progress and any person who is alive moves while standing no matter what. Cheep cameras is not a terribly bad idea. The cheapest way to scan something still is to take one good camera and go outside on a medium cloudy day(because you need A LOT of flat uniform light without shadows) and to turn the subject around on some sort of rotating stand with the camera on a tripod, take a few dozens of sharp well exposed and focused photos around all with the same camera settings. Should do 2 or 3 passes from different heights (more would be better than fewer) and then use photogrammetry to reconstruct 3d from photos of the subject with masked background. You could go around the subject as well, but it’s harder and you get a shadow - here have a look at my bean bagdone this way. If you wanted to have a cheep ‘scanner’ I think you could do with something like 48 cheep point and shoot cameras or webcams that will be easier to control, but will have less of dynamic range and crappier colors. You would look for something not less than around 8-10 megapixels and you will need loads and loads of light so it might be a good idea to get a few cheep flashes and point them to white ceiling above your subject, but you should have a plan of how you would sync them to the cameras as well as how you would trigger all the cameras at the same time. If you are thinking of continuous lighting, don’t. You would need A LOT of it and cheep LED or fluorescent bulbs do not emit all the usual light spectrum(low CRI combined with crappy cameras might not be of acceptable quality), cheep Edison bulbs while have CRI of 100 use a lot of power, because you would need really a lot of them and they emit heat. Sounds like a perfect element in a DIY project to start a fire. It sounds like a project for over a thousand euros and a lot of work anyway that’s not even counting photogrammetry software. Seems like scanning isn’t that cheep. Maybe someone here knows how one might go about it using Kinects or something…

1 Like

You can probably do a scanning rig for 500 euros if you really stretch it. Lets say 6 poles that have cameras attached on 3 levels, so 18 cameras in total. LED strip is cheap. You would have to know electronics though. Raspberry Pi Zero is 5 bucks. So without cameras that would be 18 * (5+5) = 180 euros. You can bind the Raspberrys to the LED electronics and tie them up into one single switch to take a shot at the same time. The big cost here would be the cameras. You might find IMX219 (8 megapixel) is something like 25 euros so that would increase the price by 450 already. Total of 630 euros.

Basically, if you can find good cameras for cheap, that can be locked to manual camera mode and remotely triggered at the same time, the rest is trivial. It’s all about the price.

I’ve also toyed with the idea of used mobile phones. They can be programmed and have flashes, usually with good cameras.

Once my Raspberry Zero + camera arrives, I can actually test what the quality would be on stationary objects. This package is already 66 euros as I didn’t optimize for price so it would cost ~1200 just for the programmable cameras for this very minimal rig.

This is all about capturing live subjects. If you want to capture stationary objects, you can just manually use one good camera, and that’s all you really need.

1 Like

Sounds very interesting. So I take it your plan would be to test the placement of the cameras and if everything is working with one of them and then buy more? How do they work? Would you need SD cards for every camera as well? How are they powered?

I think a google tango or 2 would be best :smiley:

(ladar + color scan)

@MartinZ: I’ve not completely thought it through yet. SD cards and power supplies would indeed increase the price.

Interesting. I could write an Objective-C / Swift app that synchronizes to other phones via Bluetooth or even simpler, a dropbox folder that contains a text file with a future timestamp in it to take a photo. All mobiles would check this folder, read the timestamp and assuming their clock’s are all millisecond synchronized, trigger the photos at the same time. The file could have multiple timestamps for additional shots.

So with 4 tripods around a subject, each tripod has 3 mobiles on it and 3 different adjustable heights with one extra for the overhead shot. 13 mobiles needed + 4 poles and clips + Rotation plate + software. Create a diffused light-room box that can be placed over a subject and uses led strips for the diffuse lighting?

I just did a dirty quick concept video of this here (assuming you rotate the rig wheel manually):
http://coopemedia.co.uk/files/mobilescan.mp4

With the wheel rig you rotate it manually, this is to keep costs down on how many cameras you need. You could use this outdoors or indoors with the tent too.

Actually, could it be even cheaper with a 180 degree arc pole that goes from left to right over a subject and it simply has wheels on it that fit into the recess groove on the ground plate. Put 5 go-pro’s (https://gopro.com/help/articles/question_answer/Controlling-Multiple-Cameras-with-One-Wi-Fi-Remote) / mobiles on it and spin that 4 times. That would probably get as much data as the 13 mobile camera solution shown above?

Doing this project with these: https://hackernoon.com/hacking-a-25-iot-camera-to-do-more-than-its-worth-41a8d4dc805c would be how I would go about it right now. Currently too many projects can’t do this one (yet)

Programmable, good image quality. Marginally cheaper (I think) than a single board computer + camera.

edit: Also this: http://janos.io/device-list.html

1 Like

3D scanning isnt something new these days, i work with kinects and realsense for robotic vision.(kinect TOF camera, RealSense= stereoscopic vision. (i’d recoment realsense only for outdoors, the overall superiour kinect doesnt work outdoors).

If you want to be on the cheap to create a 3d model of camera’s then any camera can do.
All you would need is a lot of photo’s and some smart software like : http://insight3d.sourceforge.net/

The challenge here is doing snapshot 3D scan on a shoestring budget. Something like capturing a difficult pose from a human subject can’t be easily achieved with just 1 camera. For static objects, you’re correct, 1 camera works very well.

Ah ok, your making a camera ring setup ?.
Distributed over a circle multiple camera’s take a snapthot at the same time.

Those setups are nice, they’re also used in freeze time shots. (all camera’s take the same moment but from a diffrent angle). Its not often in a circle but that allows for turning around it. Its super cool to just turn over a glass of water or so then freeze time like so, (time frozen droplets).

If you need cheap camera’s you might as well look around at the FPS quadcopter scene.
These guys use very minimalistic camera’s (due to weight limits), so often their camera’s are cheap as well.
I’ve seen 4K camera’s used there not larger then my thumb and able to record 4k for 15 minutes for (if i remember 25$ but not sure if remember price correctly). from your url i see your a thinkerer as well :slight_smile:

Cheapest (and the best) cameras are the old used mobile phone camera modules sold on aliexpress.com. 2 euros for 8k cameras. Now if there was a way to connect, trigger and transfer JPEG from those… :pensive:

http://www.arducam.com/ is also one alternative that might work.