Photos of an object -> 3D modelling -> Manufacture -> Fit the thing to the object

[Ryan .]

Hey all,

I'm wondering if anyone has succesfully executed the following process, and if so, what was the workflow?

1) Take a bunch of photos of an object (in my current case, a motorcycle, but could be anything really)

2) Use 123D catch or something similar to process the photographs into a 3D mesh

3) Design a part to against the mesh using Fusion 360 or simila

4) Print or laser cut the part, and fit it back to the original object (welding, bolts, glue, tape, whatever)

I've thought of this so many times as a great way to make rapid prototypes, exactly the soft of stuff we should be able to do with the tools we have, but I've never actually SEEN it happen!

Anyone?

R

[Ian Petrie]

a WHOLE motorcycle? Awesome!

--
You received this message because you are subscribed to the Google Groups "Reading Hackspace" group.
To unsubscribe from this group and stop receiving emails from it, send an email to reading-hacksp...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Ryan .]

Ya!!! Apparently both work now, just had the call from MTC!

On that note, anyone wanna buy a busted-ass Honda revere with an asbo exhaust and dodgy clutch?

[Stuart Livings]

On 07/07/2016 10:03, Ryan . wrote:
> On that note, anyone wanna buy a busted-ass Honda revere with an asbo
> exhaust and dodgy clutch?

Have you ever thought of going into sales or marketing as a career?

Stuart

[Matthew Daubney]

I've done it with small stuff (a razor) and a kinect to make a stand. It's... fiddly at best. Photogrammetry is proper hard.

[Ryan .]

Matt, I think I know what you mean. I've tried to get my head around it a few times, but I keep hearing it as a possibility, and to me it's a great goal. 

Even if the 3D scanning is done from something other than photos, is there a workable toolchain at all? Has anyone we know ever actually done this? Is there some other 'tool' that we need that isn't a smartphone or a kinect? 

R

--

[Tom Allen]

Very very interested in this but suspect it's only going to be easy and quick with expensive automated 3d scanning gear

[Tom Allen]

I've tried the photograph geom and although some of it was accurate, it wasn't accurate enough as a whole to use for fitted parts. I tried again with many more photos and it got worse. Less error, but in more places.

[Ryan .]

Yeah, makes sense, if the gap is with the scanner, that can be filled!

If the rest of the toolchain is workable we could beg/borrow/steal a better scanner.

[Ryan .]

Doesn't grow have something?

[Alex Gibson]

No, and yes.

At Grow there is a Fuel3D scanify.  It can take a nice, full colour 3D photograph, but in practice I have found it limited in the size of object you can scan.  We don't have the licence for the premium version allowing stitching though...

I've scanned big things with Kinects before , using Scanect. Very good results at person scale. 

Finally, I am looking for an excuse to build a proper large object scanning rig.  Malcolm, David and I gave discussed various proven options, I would go with a constellation of cameras, including raspberry Pi +cam and dslrs.  At Grow we have some meeting booths made of aluminium extrusions which are the perfect ready made rig.

Do you have a budget for this?

Cheers

Alex

Tapped on my mobile phone.

[Tom Allen]

When you say very good results do you mean aesthetically or engineering? My experience was that it was a very good likeness, but widths of key parts were off by 15% in some cases which is useless for part fitting (nominal was 3-5% but that's still way to much)

[Ryan .]

Don't give any damn what it looks like, so long as it's dimensionally accurate.

[daprigoo]

I've not seen this done as a general or high res solution although there are low res application specific examples - foot scanning and modelling and I believe body scanning for costumes, while area scanning of architecture is is a common occurence. But these are low precision examples and as Matt suggests you'd likely need an expensive industrial photogrammetry setup of largish scale high resolution/precision.

I don't believe a simple set of images will be sufficient, the mesh building algorithms do feature matching so colour and intensity plan a part and may lead to incorrect inferences.

The Kinect V1 scanner uses two cameras which capture an RGB image and a depth "image" respectively. The depth data is used to craete the mesh which can then be dressed with the RGB image as a skin, here is an example using Skanect (left is 3D and skin, mid top is the r/t RGB view, top right is r/t depth view, bottom right is r/t scanner model (ie built from depth info) view;

CAT scanning is probably the closest to providing very accurate scan models, but that won't work for all materials (ieg not metal) and there are restrictions on size.

[daprigoo]

I think the most likely approach will be a structured light system, the Kinect uses structured IR for it's depth camera but you'd want something much more accurate and that means precision alighnment and optics (ie high cost)

These claim quite high resolution (to +/- 0.02mm) - http://www.laserdesign.com/products/rexcan-cs-3d-scanner/

What fabrication technology were you contemplating as that will influence the resolution you need?

[Ryan .]

It'd be easier to show that tell, tbh, but one thing I'd like to try is laser cutting plywood templates to guide the plasma cutter, with machining or grinding to finish.

Thinking 1-2mm accuracy generally is fine.

[Alex Gibson]

I have a CNC router platform based on Openbuilds OX, sized 1m x 750mm.

Would you like to try running the plasma cutter on this?

I already looked at making a simple 3D printed mount for it, that I can do quickly and easily, I haven't started only because I have other calls on time.  If you fancy doing the commissioning I can loan you the bulk of the machine ready to go! 

Tapped on my mobile phone.

On 7 Jul 2016, at 13:13, "Ryan ." <ry.w...@gmail.com> wrote:

It'd be easier to show that tell, tbh, but one thing I'd like to try is laser cutting plywood templates to guide the plasma cutter, with machining or grinding to finish.

Thinking 1-2mm accuracy generally is fine.

On 7 Jul 2016 1:12 p.m., "'daprigoo' via Reading Hackspace" <reading-...@googlegroups.com> wrote:

I think the most likely approach will be a structured light system, the Kinect uses structured IR for it's depth camera but you'd want something much more accurate and that means precision alighnment and optics (ie high cost)

These claim quite high resolution (to +/- 0.02mm) - http://www.laserdesign.com/products/rexcan-cs-3d-scanner/

What fabrication technology were you contemplating as that will influence the resolution you need?

On 07/07/2016 13:05, daprigoo wrote:

I've not seen this done as a general or high res solution although there are low res application specific examples - foot scanning and modelling and I believe body scanning for costumes, while area scanning of architecture is is a common occurence. But these are low precision examples and as Matt suggests you'd likely need an expensive industrial photogrammetry setup of largish scale high resolution/precision.

I don't believe a simple set of images will be sufficient, the mesh building algorithms do feature matching so colour and intensity plan a part and may lead to incorrect inferences.

The Kinect V1 scanner uses two cameras which capture an RGB image and a depth "image" respectively. The depth data is used to craete the mesh which can then be dressed with the RGB image as a skin, here is an example using Skanect (left is 3D and skin, mid top is the r/t RGB view, top right is r/t depth view, bottom right is r/t scanner model (ie built from depth info) view;

<mime-attachment.jpg>

CAT scanning is probably the closest to providing very accurate scan models, but that won't work for all materials (ieg not metal) and there are restrictions on size.

On 07/07/2016 09:47, Ryan . wrote:

Hey all,

I'm wondering if anyone has succesfully executed the following process, and if so, what was the workflow?

1) Take a bunch of photos of an object (in my current case, a motorcycle, but could be anything really)

2) Use 123D catch or something similar to process the photographs into a 3D mesh

3) Design a part to against the mesh using Fusion 360 or simila

4) Print or laser cut the part, and fit it back to the original object (welding, bolts, glue, tape, whatever)

I've thought of this so many times as a great way to make rapid prototypes, exactly the soft of stuff we should be able to do with the tools we have, but I've never actually SEEN it happen!

Anyone?

R

--
You received this message because you are subscribed to the Google Groups "Reading Hackspace" group.
To unsubscribe from this group and stop receiving emails from it, send an email to reading-hacksp...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "Reading Hackspace" group.
To unsubscribe from this group and stop receiving emails from it, send an email to reading-hacksp...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Ryan .]

Na, thanks, bit big, and I think the laser cut template thing would make nice work of it.

Let me try that first!

[daprigoo]

We could try an experiment with a Kinect v1 and Skanect to create a model then measure & scale. Bear in mind the scanner uses light so shiny surfaces don't work. You also need a spiny backless stool or platform as I've found the best results come from keeping the scanner still (on a tripod) and moving the subject, eg;

[Stuart Ward]

On 7 July 2016 at 09:47, Ryan . <ry.w...@gmail.com> wrote:

1) Take a bunch of photos of an object (in my current case, a motorcycle, but could be anything really)

2) Use 123D catch or something similar to process the photographs into a 3D mesh

3) Design a part to against the mesh using Fusion 360 or simila

4) Print or laser cut the part, and fit it back to the original object (welding, bolts, glue, tape, whatever)

I have done the first 3 steps, took a bunch of photos of both the Queen, and Fortanbury Lion, 123Catch to create the mesh, edited the mesh in meshlab, but haven't printed them as what came out was a blob vaguely shaped like the original. 

-- Stuart Ward M +44 7782325143

[Malcolm Napier]

Just to add to/reinforce what David has said, he and I have successfully done people with Kinect V1 for artistic purposes - including the scan for both attempts at 3d2ring - the heads from the statue of Alan Turing.

123D-Catch was a bust (pun intended) for 3d2ring. I ended up buying a Kinect and a motor cycle battery to create a portable scanner. Luckily I needed a new laptop at the time - so the quad core i7 with Kinect compatible graphics card didn't need to be charged against the project. Since then I have not gone anywhere near 3d photographic scanning. I spell this out to illustrate how far off the mark 123D-Catch was

Down the line, I am interested in how dimensionally accurate scans of people are for three purpose of clothes hacking. (It is a long story but this was a major factor in Hannah and me getting involved in 3d printing originally.)

If there are plans to experiment with the accuracy of scanning, then I am interested but probably not able to contribute much time wise in the near future.

[Ryan .]

So the hole is in the scanning tech still?

How to get around this? What's the best 3D scanner for mechanical parts money can buy? One of these Faro things?

What about one of their wavey-pointy-multiple-joint-stick-arm things with a probe on the end you just poke at your object?

R

[Hugo Mills]

On Thu, Jul 07, 2016 at 03:07:24PM +0100, Ryan wrote:
> So the hole is in the scanning tech still?
>
> How to get around this? What's the best 3D scanner for mechanical parts
> money can buy? One of these Faro things?
>
> What about one of their wavey-pointy-multiple-joint-stick-arm things with a
> probe on the end you just poke at your object?

That might actually be pretty easy to build the hardware: a bunch
of rotary encoders on a jointed arm. Converting a set of encoder
values to a position is just a trivial bit of trigonometry/matrix
maths. From that point, the issue is turning the resulting point cloud
into a mesh, but there seem to be a few things out there that will do
that part already.

.oO( I wonder if there's any benefit in a post-processing step
where you indicate a part of the surface and specify that it's part of
a geometric shape, like a cylinder or a cone or a plane, and it looks
for appropriate solutions )

At least with physical contact on the surface, you're going to have
fewer problems with accuracy of the point locations (assuming your
rotary encoders have enough resolution, and the joints have minimal
backlash(*)).

Finding an arm design that gives you the maximum precision for the
widest range of positions could be an interesting challenge -- you
want to avoid regions where large changes of position map to small
changes of angles (kind of like gimbal lock in reverse).

Hugo.

(*) OK, big assumption.

> R
>
> On 7 Jul 2016 3:02 p.m., "Malcolm Napier" <malcol...@gmail.com> wrote:
>
> Just to add to/reinforce what David has said, he and I have successfully
> done people with Kinect V1 for artistic purposes - including the scan for
> both attempts at 3d2ring - the heads from the statue of Alan Turing.
>
> 123D-Catch was a bust (pun intended) for 3d2ring. I ended up buying a
> Kinect and a motor cycle battery to create a portable scanner. Luckily I
> needed a new laptop at the time - so the quad core i7 with Kinect
> compatible graphics card didn't need to be charged against the project.
> Since then I have not gone anywhere near 3d photographic scanning. I spell
> this out to illustrate how far off the mark 123D-Catch was
>
> Down the line, I am interested in how dimensionally accurate scans of
> people are for three purpose of clothes hacking. (It is a long story but
> this was a major factor in Hannah and me getting involved in 3d printing
> originally.)
>
> If there are plans to experiment with the accuracy of scanning, then I am
> interested but probably not able to contribute much time wise in the near
> future.
>

--

Hugo Mills | UNIX: Italian pen maker
hugo@... carfax.org.uk |
http://carfax.org.uk/ |
PGP: E2AB1DE4 |

[Matthew Daubney]

Ring up Creat3d and ask if you can test their range of scanners. They carry a couple of commercial models from £xxx to £xxxxx type prices.

[Ryan .]

Exactly what I needed! Good call Matt! Do we know anyone there?

[Matthew Daubney]

No. But I've just turned up before and chatted to them. They seem nice enough.

[Alex Gibson]

Yep.  Great bunch.  Simon, Jon and Sabina.  Let me know if you want an introduction.

Tapped on my mobile phone.

[Ryan .]

Thanks all, I spoke to Simon on the phone and chatted through the requirement. 

He's echoed your general understanding of 'it's not really that good for mechanical stuff yet' but was ready to demonstrate it to me so I could try it. 

I'm going to try to go down and have a play with some of their kit. They recommended the use of David SLS for scanning, and suggested that the shiny chrome bits could be coated with a wash-off white matte spray that David vision systems supply for the purpose. http://www.desktop3dprinter.com/accessories-spares/david-coating-spray.html

Has anyone used David before?? I seem to remember hearing about it in early days...and may have even tried it! It sounds like something we'd be able to build a system around at the space, with a good USB3 camera and a decent DVI projector as the key hardware requirements (from what I can see). 

What do you think?

R

[mikethebee]

Looking through the posts for the podcast, I was reminded of the scanner that we saw at the open day at Harwell in the engineering dept. That might be available if you ask them nicely. 

An alternative for dimensional accuracy it a classic CMM, Co-ordinate Measuring Machine, though I don't have access to one myself anymore. 

-MtB

On Friday, 8 July 2016 09:46:32 UTC+1, Ryan wrote:

[Tom Allen]

For some cases I think a co-ordinate measuring arm would work great. For me some cases would be as simple as running it over a few curved faces a few times to get a plane and that would be enough, I don't need to scan the hole things, just the two or three mounting points i want to attach a new part to.

I bet they aren't cheap in the .1 mm accuracy range though?

[Chris Ward]

A delta format 3D printer with no base would probably work quite well. A foot on the bottom of each post would rest on the object to be scanned, and the hot end would be replaced with a simple probe to detect contact. With no base on the printer, the probe can even cope with concave surfaces which are below the level of the feet.

The auto bed leveling software could be re purposed to scan the surface at the required X-Y resolution. 

[Ryan .]

Some progress:

Autodesk Memento/Remake - Tried it out, not brilliant. Could likely be improved with practice and better photography?

Also, I've picked up a newer kinect+adapter from Alex, as well as a fairly nice webcam. 

I think the next step is to build as beefy a PC as possible, dedicated to the task, dual booting with an nVidia graphics card. 

From the other thread, we should then install this software on it:

http://mesh.brown.edu/calibration/software.html

Updated in 2015, includes source code, mac and windows binaries, one projector and one webcam (directshow?)

http://www.3dunderworld.org/software/

Updated july 11th, no installation instructions, seems to be windows only, depends on openCV and Canon cameras, 

https://github.com/jakobwilm/slstudio

Targetted to real-time stuff. Not ideal for mechanical parts

https://code.google.com/archive/p/structured-light/downloads

Old, think I've tried it before, sucked then? 

The only bit of hardware missing is a Canon camera!

[Lonn Brown]

I can bring mine to the lab for a few hours - when would be a good time to meet?

[Malcolm Napier]

Are we getting two things intertwined here:

1) Structured light scanning
2) Kinect Scanning

Both David and I have laptops powerful enough for Kinect Scanning (Mine is Quad core i7, 12GB RAM, NVidia 540M Graphics card. David's is similar/more modern/has USB3 for Kinect 2 scanning). Between us we have done quite a lot of Kinect scanning - mainly of people. David has even designed a 3d printable handle to make the scanner hand-held. I have a motorcycle battery to make it portable (although the scanning hammers the laptop battery if used in this mode).

How much beefier do you need the PC to be for structured light scanning? I.e. couldn't we use my laptop? I have spare SSD drives and a plug- in bay on my laptop, so that we can have a dedicated build. And I have other laptops for daily use.

In the meantime, if I don't get distracted by all the other things going on at EMF, I will try some Kinect scanning of non-human (engineering) objects and assess for dimensional accuracy across a range of sizes from phone sized to body panel sized.

Regards,
Malcolm

[Ryan .]

Kinect or structured light, I'm planning on getting set up to try both and see which works best for what.

[Ryan .]

Hi Lonn (and anyone else with a canon camera)

Are you around this weekend to try this out?

R

[Mr E]

Hi Ryan,

I was thinking about this.  And I did notice that in Windows 10, there's an Microsoft 3D scanning app, which can feed into their 3D Builder app, which can export a STL file.  I've no idea if it's any good, but if I can get my hands on a Kinect V2 (I believe it's the one with the full "Time Of Flight" camera), I will certainly give it a go.

I'm sure being MS it will have limits, but it does occur to me, that it's likely to be moderately hacker proof, and will work after a fashion.

https://developer.microsoft.com/en-us/windows/hardware/3d-scanning-with-kinect

Rupert

To unsubscribe from this group and stop receiving emails from it, send an email to reading-hackspace+unsubscribe@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "Reading Hackspace" group.

To unsubscribe from this group and stop receiving emails from it, send an email to reading-hackspace+unsubscribe@googlegroups.com.

[Ryan .]

I think I've got a kinect V2, is that the square one? Alex lent it to me to play with this. I've also got the USB connector thing for it.

R

[Mr E]

That sounds right :)

The requirements are a Win10, a reasonable Nvidia GPU, *must* have Intel or Renesas USB 3.0 (it pushes quite a lot of data) and a Kinect v2.

To unsubscribe from this group and stop receiving emails from it, send an email to reading-hackspace+unsubscribe@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "Reading Hackspace" group.

To unsubscribe from this group and stop receiving emails from it, send an email to reading-hackspace+unsubscribe@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "Reading Hackspace" group.

To unsubscribe from this group and stop receiving emails from it, send an email to reading-hackspace+unsubscribe@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "Reading Hackspace" group.

To unsubscribe from this group and stop receiving emails from it, send an email to reading-hackspace+unsubscribe@googlegroups.com.

[Mr E]

Should be this one - https://en.wikipedia.org/wiki/File:Xbox-One-Kinect.jpg

[Ryan .]

Hmm... Does anyone have a spare computer with the right USB chipset for a Kinect 2 and room for an nVidia GPU? That or a PCIe to USB3 card with the right USB3 chipset?

On Fri, Aug 5, 2016 at 12:05 PM, Mr E <excha...@gmail.com> wrote:

[Mr E]

What were the spec's of that one we were trying to get Ubuntu + VirtualBox + pass through working on?

To unsubscribe from this group and stop receiving emails from it, send an email to reading-hackspace+unsubscribe@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "Reading Hackspace" group.

To unsubscribe from this group and stop receiving emails from it, send an email to reading-hackspace+unsubscribe@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "Reading Hackspace" group.

To unsubscribe from this group and stop receiving emails from it, send an email to reading-hackspace+unsubscribe@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "Reading Hackspace" group.

To unsubscribe from this group and stop receiving emails from it, send an email to reading-hackspace+unsubscribe@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "Reading Hackspace" group.

To unsubscribe from this group and stop receiving emails from it, send an email to reading-hackspace+unsubscribe@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "Reading Hackspace" group.

To unsubscribe from this group and stop receiving emails from it, send an email to reading-hackspace+unsubscribe@googlegroups.com.

[Ryan .]

No USB3 on the motherboard, and the USB card I don't think works. A USB PCIe card would work in it. 

[Mr E]

Ah ok.  My x230 laptop is i7, SSD and Intel USB 3.0, but just not nVidia GPU.  I've been pondering getting the external expresscard <--> PCIe adaptor for it, for gaming.  Maybe this just moved up the list :)

But if anyone else has got a fairly new i7 Intel PC, it'll probably be quicker.