MagWeb config file settings for 2.5 lens test?

[Darnell Williams]

Any chance of posting config settings MagWeb used for his experiments with
2.5 lens adaptations?

[Christoph Heindl]

Am Montag, 19. März 2012 18:12:26 UTC+1 schrieb Darnell Williams:

Any chance of posting config settings MagWeb used for his experiments with
2.5 lens adaptations?

Only MagWeb knows the real config, but from what he explained in other threads I will provide a quick guess

camera_size_x: 640

camera_size_y: 480

camera_fx: 514.16

camera_fy: 514.16

camera_px: 320

camera_py: 240

camera_near: 100

camera_far: 2000

volume_size: 512

volume_min {

  x: -250

  y: -250

  z: 400

}

volume_max {

  x: 250

  y: 250

  z: 900

}

integrate_truncation: 10

integrate_max_weight: 64

icp_max_iter: 20

icp_max_dist2: 200

icp_min_cos_angle: 0.9

smooth_normals: false

disable_optimizations: false

extract_step_fact: 0.5

Note that the focal lengths (fx,fy) are not only affected by the dioptre of the glasses used, but also by the mounting 'accuracy'. Especially the distance of the glasses to the sensor plays an important part. Therefore your values will have to be slightly different. I would recommend to perform a chessboard calibration. I might come up with a simple tool to generate the required images and invoke the opencv calibration.

Best,

Christoph

 

[MagWeb]

Right guess :) exactly my settings!  - Sorry for being so late.

My lenses were as close as possible to the Kinect's housing. But for they are bowed (as glasses are) your distance might be different (caused by the size of the lenses).

Meanwhile I use the lenses separated from each other to increase their distance to each other to have they centered on the projector and the sensor (this cuts my RGB cam but I can see enough for orientation) and fixed them with tape.

Gonna order some better lenses and build a stable housing....

[Darnell Williams]

Thanks to you both. This is incredibly fun.

[Darnell Williams]

My first scan using MagWeb's lens hack. Cool.

[Christoph Heindl]

Am Mittwoch, 21. März 2012 09:00:36 UTC+1 schrieb Darnell Williams:

 My first scan using MagWeb's lens hack. Cool.

Nice! Is that full 360°

Best,

Christoph 

[Mark Schafer]

Woo hoo. Go Darnell :-)

[Darnell Williams]

It's a 360 scan. Bit distorted. I spun myself about on a chair and kept the camera static.
Thanks Mark. ;). Want to do lots more.

[approx]

On Wednesday, March 21, 2012 5:59:41 PM UTC-7, Darnell Williams wrote:
> It's a 360 scan. Bit distorted. I spun myself about on a chair and kept the camera static.

> Thanks Mark. ;). Want to do lots more.
>

> On Wednesday, March 21, 2012 4:15:16 AM UTC-7, Mark Schafer wrote:<blockquote class="gmail_quote" style="margin:0;margin-left:0.8ex;border-left:1px #ccc solid;padding-left:1ex">Woo hoo. Go Darnell :-)
> </blockquote>

Hi Darnell,
Sorry I am new to scanning, so maybe it is a silly question..but what is the benefit of using the 2.5 lens? Does it increase the resolution of the scan? Also by reducing the focal length value in the config file, does that mean you need to be closer to the camera to take the scan? how far were you from the camera?

[Darnell Williams]

I'm new to scanning myself. The 2.5 "magnifying" reading lens allow for greater surface detail.
The focal length must be changed to account for the lens, not the volume defined. But you can get
a bit closer, just define your volume accordingly.

[jrde...@gmail.com]

Been following this for a little bit, I just picked up a kinect yesterday so I'll probably get started in the next few days.

Read through a few posts and using the lenses seems like a great idea... I saw this in best buy, has anyone tried this?..Not sure what exactly it does but might have a similar effect as the glasses..

http://www.amazon.com/Zoom-Kinect-Xbox-360/dp/B0050SYS5A

-James-

[Misha]

Hey, that's my kinect Darnell's using.... I just had to jump in.... 

My question is what is the command that utilizes the cfg script? 

I usually start the ap with a double click, using : 

Start ReconstructMe Realtime OpenNI 

...and answer y when prompted to agree to license terms...is this 

where i source the cfg file for the lens? what is the command in the 

shell? 

This is a great program and I know Darnell and I are really excited 

about the potential.... 

Here's a couple of my scans, using reconstructme...

Cheers, 

Mike 

[Misha]

Never mind...I got the cfg working...thanks... here's the results....added a little color...

Cheers,

Mike

[Bill Culverhouse]

How are you doing the texture mapping?

-b

[Darnell Williams]

Taking photos and using Zbrush "Spotlight".

[Darnell Williams]

Nope. As stated in earlier posts, the Nyko widens the view to include more area, you want to magnify your view
with the reading lens.The closer you are, the better the detail.

[Misha]

Thanks Darnell.... :)

Cheers,

Mike

[Misha]

and ditto on the thanks to Christoph and Magweb.... :)

[Milazi HM]

hi guys, i wondering what's kind of glasses is ok ,myopic lenses or
presbyopic glasses ?thanks .

> >http://www.amazon.com/Zoom-​Kinect-Xbox-360/dp/B0050SYS5A<http://www.amazon.com/Zoom-Kinect-Xbox-360/dp/B0050SYS5A>
>
> > -James-

[kenwshmt]

using the posted config.

this is somewhat more useful with the reading glasses.

this picture was after 10 minutes of tinkering.. admittedly not in an environment best suited for digitizing. I havnt gotten an adequate half or 360 scan yet from this. I seem to have a bilateral symmetry problem, it just wont read past the center point. oddly enough, it will happily scan the back of my head for 300 degrees, just not across my face. I may just do a left right mirror if I cant resolve it.

the was a $8 pair of 2.5 glasses from wallmart... style #M10188SA    sku 715020083 2
it is sufficiently large to reach the bottom of the laser and the cameras. The video camera has the frame down the middle, but it is still usable for framing.
I use a glasses neck strap attached to the stems to tie it behind the kinect, to keep it in place.

[Tony Buser]

Here's my results using +2.5 glasses:

http://www.flickr.com/photos/tbuser/7013251283 

http://www.flickr.com/photos/tbuser/7013251877

I'm also having trouble getting a full 360 degree scan.  But it's very close.  Great improvement.  :)

[Christoph Heindl]

Fantastic!

As far as your setup is concerned, it looks like you might get better result if the IR projector and the IR camera look through the center of the glasses to reduce distortion effects.

I'm considering a simple calibration tool to access the correct settings.

Best,

Christoph

[Darnell Williams]

Nice test case sample.

[Bill Culverhouse]

I too have had trouble finding glasses wide enough to get the cameras looking through
the center of the lens. Or at least the less distorted non-center.

That said, Tony,  excellent choice for frames though. Nerdy Kinect FTW.

Is there no one that knows enough about optics to say which lenses one should buy
from Edmund Scientific (or whoever). Come up with a 3d printable "Frames" for them
and there you go.

A calibration tool would be very nice.

-b

[kenwshmt]

[Bill Culverhouse]

Wow, be careful with those babies. Looks like you are about to burn a couple of holes in that
back wall. :-)

-b

[Christoph Heindl]

Am Sonntag, 25. März 2012 14:22:46 UTC+2 schrieb Bill Culverhouse:

I too have had trouble finding glasses wide enough to get the cameras looking through
the center of the lens. Or at least the less distorted non-center.

Why not breaking it, so you are left with two separate glasses that can then be attached individually?

 

A calibration tool would be very nice.

I have started to work on that :) I'd like to see how far this idea can be pushed (is 2.5+ the maximum?). Therefore I will provide the community with a calibration tool. I hope a lot of you will pick that up and put something together (tutorial ? :) for rookies. Note that this tool will calibrate the intrinsics of the depth sensor, the pattern distortion cannot be calibrated. Also note, that this will only work with OpenNI, since only OpenNI allows direct access to IR mode which is required for this kind of calibration.

Best,

Christoph

[MagWeb]

Am Sonntag, 25. März 2012 18:07:19 UTC+2 schrieb Christoph Heindl:

Why not breaking it, so you are left with two separate glasses that can then be attached individually?

 

Yes, that´s what I meanwhile did - it may disturb your RGB cam, but that doesn´t matter. Definitely better results 

I have started to work on that :) I'd like to see how far this idea can be pushed (is 2.5+ the maximum?). Therefore I will provide the community with a calibration tool. I hope a lot of you will pick that up and put something together (tutorial ? :) for rookies. Note that this tool will calibrate the intrinsics of the depth sensor, the pattern distortion cannot be calibrated. Also note, that this will only work with OpenNI, since only OpenNI allows direct access to IR mode which is required for this kind of calibration.

 

Great , I hope to get a second set of  lenses soon ( such as an eye-doctor uses to set the glasses strength) - so I will have variable pairs of lenses. I´ll keep you informed about my results. 

[kenwshmt]

can someone with an adequately sensitive camera in the IR range take a picture of a close object in the ideal range with and without the lenses?

On Monday, March 19, 2012 12:12:26 PM UTC-5, Darnell Williams wrote:

Any chance of posting config settings MagWeb used for his experiments with
2.5 lens adaptations?

[Milazi HM]

Hi i'm in trouble to understand the "+2.5 glasses " exactly  means in english culture . is it kind of glasses for myopic eye ?

[Misha]

Ready made magnifying lenses for reading glasses are available in strengths ranging from 1.00 to 3.50. The glasses are readily available in many stores without prescription.

Cheers,

Mike 

[Tony Buser]

Since I'm impatient, I tried to figure out how to do the chessboard calibration using RGBDemo.  I think I got it.  However, what I don't understand is how to turn it's findings into camera_fx/fy numbers?  Here's what I got using +2.5 lenses:

depth_intrinsics: !!opencv-matrix
   rows: 3
   cols: 3
   dt: d
   data: [ 5.4143834003803727e+02, 0., 3.2793017195129437e+02, 0.,
       5.4263420055187089e+02, 2.5541469754592663e+02, 0., 0., 1. ]

As a wild guess, do I take the first number in each line and multiply by 100?  So in my case, fx = 541.43 and fy = 542.63 ?

Also today I found a local dollar store sells reading glasses for... $1 each  :) So I picked up a few different kinds (+3.0 and +3.5) that I'll play with once I figure out how to calibrate this stuff.

[Milazi HM]

Thanks Mike, i thinke i'll buy this from a glasses store easily.

Best,

Milazi

--
Milazi Hm

Twitter:  @milazi

Website:    milazi.com

[Christoph Heindl]

Tony, 

Am Montag, 26. März 2012 06:37:51 UTC+2 schrieb Tony Buser:

Since I'm impatient, I tried to figure out how to do the chessboard calibration using RGBDemo.  I think I got it.  However, what I don't understand is how to turn it's findings into camera_fx/fy numbers?  Here's what I got using +2.5 lenses:

depth_intrinsics: !!opencv-matrix
   rows: 3
   cols: 3
   dt: d
   data: [ 5.4143834003803727e+02, 0., 3.2793017195129437e+02, 0.,
       5.4263420055187089e+02, 2.5541469754592663e+02, 0., 0., 1. ]

As a wild guess, do I take the first number in each line and multiply by 100?  So in my case, fx = 541.43 and fy = 542.63 ?

Yes that's true it's 5.41*10^2. But watch out, RGBDemo aligns the RGB sensor with the Depth sensor ( see the intrinsics of the RGB camera they should be the same). Therefore this calibration cannot be used with ReconstructMe. ReconstructMe needs a depth-only calibration. I'm working on a calibration that works for ReconstructMe.

[Tony Buser]

Oh so you can't use RGBDemo to do it?  How do you do a chessboard calibration with only the depth sensor?  Don't you need the RGB cam to see the squares?

Thanks!  Will reconstructme errors be reduced/360 degree scans be easier once it's properly calibrated?  Even if the scan volume needs to be reduced to a tiny box, I'd be super happy to be able to get high res scans of small objects out of this.

[Christoph Heindl]

Am Montag, 26. März 2012 07:57:18 UTC+2 schrieb Tony Buser:

Oh so you can't use RGBDemo to do it?  How do you do a chessboard calibration with only the depth sensor?  Don't you need the RGB cam to see the squares?

You could, but source code must be changed. The calibration is done using the depth sensor, which is an infrared sensor. The chessboard is visible in infrared.

 

Thanks!  Will reconstructme errors be reduced/360 degree scans be easier once it's properly calibrated?  Even if the scan volume needs to be reduced to a tiny box, I'd be super happy to be able to get high res scans of small objects out of this.

I hope so but I don't know how the glasses distort the infrared projector pattern. The calibration only helps with the depth sensor intrinsics. 

[Rafael Daniel Zuban]

Hi to all :) I tried attach today a 3.0 dip glass, to make a test with MagWeb set - but with this settings it makes nothing with my glasses.

- This lenses are a bit thick, and are from plastic - can be this problem?

- put the lenses out from frame, the reult is same

thanks,

Raff

2012. március 19., hétfő 18:12:26 UTC+1 időpontban Darnell Williams a következőt írta:

Any chance of posting config settings MagWeb used for his experiments with
2.5 lens adaptations?

[Christoph Heindl]

Am Montag, 26. März 2012 18:41:14 UTC+2 schrieb Rafael Daniel Zuban:

Hi to all :) I tried attach today a 3.0 dip glass, to make a test with MagWeb set - but with this settings it makes nothing with my glasses.

Just an idea: are those + or - 3 dip? 

[Eric McCann]

I tested 3.5d glasses. They are not working at all. Also, if I take one lense of 2.5d, and reverse it, (Careful not to scratch the lense on the kinect) I seem to get better results. So, the depth cam I have the lens connected like regular glasses, on the IR Lasers, I have the lens flipped. Messing with the calibration settings, doesn't seem to make a difference good or bad for me. It would help if I knew what the other settings meant....

integrate_truncation:
integrate_max_weight:
icp_max_iter:
icp_max_dist2:
icp_min_cos_angle:
smooth_normals: (I am certain of what this is.)
disable_optimizations:(Not precisely sure)
extract_step_fact:

[Rafael Daniel Zuban]

its a +3 dip, maybe the lens is not good. i will try with another lenses...

[Rafael Daniel Zuban]

Maybe i can test it with 2.5 dip lenses, thanks. i think the shape is also important. this common reading glasses are cheap (and maybe not so perfect) like professional optics.

[Mark Schafer]

For those of you who would like to experiment with some lenses of the cheap variety but better than distorted reading glasses. (its hard to position the optical centers so they match - which causes distortion).
Try these:
DCX lens (50 assorted) $7 - http://www.sciplus.com/singleItem.cfm/terms/14448
Classy magnifiers (6 mag=2x) $15 - http://www.sciplus.com/singleItem.cfm/terms/10818
Loupe (mag=4x) buy2 $12 - http://www.sciplus.com/singleItem.cfm/terms/16501
Doublet(achromat) (mag=5x) buy2 $6 - http://www.sciplus.com/singleItem.cfm/terms/16846
or possibly best(easy mount) :
Twin loupe (mag=3x) $9 - http://www.sciplus.com/singleItem.cfm/terms/8939

Of course you can waste a lot of time on this site :-)
http://www.sciplus.com/category.cfm/subsection/17

[Andrew szurley]

Have any of these been successful?

Thanks for the link!

a

[Gmail - neonmark]

The most successful were the stor bought +2.5 magnification reading
glasses. Largish lenses. Make sure to center the lens over the two
kinect elements