Dear DAVID Community,

As you are probably aware, the entire DAVID company, team and product range has been integrated into HP Inc.
This forum is now read-only. Relevant content has been migrated to HP and merged into HP's Support Forums on November 1, 2016.

To start new discussions for 3D Scanning, please register and post your new topic at the HP Support Forums

Achievable absolute accuracy of DAVID

The place for questions, problems, comments and tips regarding the camera calibration.

Achievable absolute accuracy of DAVID

Postby WalterMo » Tue Apr 20, 2010 1:43 pm

Khalid Khattak suggested to open a new thread where we can show what absolute accuracy is achievable by DAVID.
But we cannot say the accuracy is e.g. 0.1mm. It depends on the object size and the relation of object size to the used calibration sheets and how the object fits to the camera image.

So let me start with an old planeless scan of the backside of my car:
viewtopic.php?p=7962#p7962

By means of a tool of MeshLab the width of the numberplate was measured to 516mm. In real it is 520mm. Measurement of the scan was 0.8 % too little.

The width of the Toyota emblem in the same scan was measured to 108.5 mm. In real it is 108mm. Measurement of the scan was 0.5 % too much.

The scale length of the used „old“ calibration sheets was 99mm.

Or do you have a better / another idea to determine the accuracy? Comments are as always welcome.

Walter
Attachments
Messung Nummernschild.jpg
User avatar
WalterMo
Moderator
 
Posts: 2363
Joined: Mon Apr 02, 2007 6:52 pm
Location: Braunschweig, Germany

Re: Achievable absolute accuracy of DAVID

Postby WalterMo » Tue Apr 20, 2010 3:26 pm

And here we have something smaller. It's the backside of a commemorative coin in honour of Kopernikus.
viewtopic.php?p=10651#p10651

It was a planeless scan with the Logitech Pro9001 at a resolution of 1600 x 1200 pixels. Scale length of the old type of calibration sheets was 8mm. The real diameter of the coin is 29mm. Now measured by MeshLab = 28.8mm. That means the measurement is 0.7% too little.

Walter
Attachments
Rueckseite mit Messung.jpg
User avatar
WalterMo
Moderator
 
Posts: 2363
Joined: Mon Apr 02, 2007 6:52 pm
Location: Braunschweig, Germany

Re: Achievable absolute accuracy of DAVID

Postby MagWeb » Tue Apr 20, 2010 4:23 pm

Hi Walter,

I´m not sure about the ML measurement

I just checked to measure the diagonal of some primitive flat square made in some 3D software.
In ML I get 423.183 (I tried several times to hit the edges)
in Rapidform explorer I get 424.264

Think the hardest part to get right measurements in ML is to catch the right points for there is no detection of edges.
So the reason why you get a shorter measurement may be caused by ML?

EDIT I know the mesh being 300x300
Measuring the length of one side I get only 299.4
Wherever there´s a surface you can set a point for the measurement (no snap on edges or vertices)
MagWeb
Moderator
 
Posts: 2499
Joined: Wed Jul 18, 2007 8:48 pm

Postby hal » Tue Apr 20, 2010 4:36 pm

Hello Walter,

I think that yours way to show the achievable accuracy is incisive.
I hope to contribute to this thread with the scan of a real scooter, on which locate some significant little parts/portions.
The raw idea is to scan with different setups and scale factors, and after join all in an unique mesh.
The final goal is to have a scan with differents LOD (Levels Of Detail) and measure every pieces, big and little.

Walter, in your measure I read a very little % of error. Maybe the causes could became from:
- a little error of the scanner;
- the difficulties to measure correctly the size reference line on the markers sheets (with a common ruler);
- the difficulties to measure the final mesh, because not ever is easy to find the right/same points of measure in the rela and scanned mesh.

To me yours percentages of error seems acceptable.

Mattia
User avatar
hal
Moderator
 
Posts: 1527
Joined: Thu Mar 01, 2007 4:08 pm
Location: Cavaria con Premezzo (VA), Italy.

Re: Achievable absolute accuracy of DAVID

Postby Khalid Khattak » Tue Apr 20, 2010 4:42 pm

Hi Walter,
First of All i am very thankful to you for giving us information about David Scan accuracy. I am really feel so proud for making my wish realistic...:)
Now, as Gunter said, the MeshLab is not a good tool to judge the accuracy of scanned model..This is because i have also used the meshlab and there is a hell of difference between Meshlab and Rapidform results..

As for as i know Rapidform Explorer is a free utility and we can work it out..

I am happy because your camera is the same as of mine and this is plausable for all the persons who want to know about the scan accuracy for this type of camera...

I hope many of us including Gunter, Greg and Mattia will put their pictorial results and make us pure belief on David..

The accuracy question comes into my mind when i checked the scan results of my setup and they were all large from the original object.. In the last turnaround of my fertilizer plant, i was having the oppurtunity to scan Turbine Casing and compare it with the install one but i lost... I afraid that i can not get the good accuracy from David as per my previous bad experiences..But this thread is prooving that my Setup was wrong, my calibration panel scale setting may be wrong or anything else..

I hope most of us will take part in this discussion and many of us together as a TEAM will proove to the world what we can achive with our beloved low cost David...

Regards and please keep posting.
PC: Dell Inspiron i3, 2.27GHz,4GB RAM,64 bit windows 7
CAMERA:
David CCD Mono USB
LASER:
David 5mW Green line laser
PROJECTOR:
3D LED Mini Projector 250lumens, Native 1024x600
[b]http://free3dscans.blogspot.com/
Khalid Khattak
 
Posts: 942
Joined: Thu Apr 23, 2009 1:30 pm
Location: Pakistan

Re: Achievable absolute accurary of DAVID

Postby Khalid Khattak » Tue Apr 20, 2010 4:50 pm

MagWeb wrote:Hi Walter,

I´m not sure about the ML measurement

I just checked to measure the diagonal of some primitive flat square made in some 3D software.
In ML I get 423.183 (I tried several times to hit the edges)
in Rapidform explorer I get 424.264


Hi Walter, What was the Original Length of diagonal?
Regards
PC: Dell Inspiron i3, 2.27GHz,4GB RAM,64 bit windows 7
CAMERA:
David CCD Mono USB
LASER:
David 5mW Green line laser
PROJECTOR:
3D LED Mini Projector 250lumens, Native 1024x600
[b]http://free3dscans.blogspot.com/
Khalid Khattak
 
Posts: 942
Joined: Thu Apr 23, 2009 1:30 pm
Location: Pakistan

Re: Achievable absolute accuracy of DAVID

Postby WalterMo » Tue Apr 20, 2010 5:35 pm

Khalid Khattak, I think you meant Gunter.
But let me answer, perhaps I am right. :wink:
The original diagonal length is, assuming the square was precisely made: (root of 2) x 299.4 = 423.416

Gunter, when we compare the measurements of the square item done by MeshLab and Rapidform explorer the difference is only 0.255%. For a first test it should be sufficient.

Walter
User avatar
WalterMo
Moderator
 
Posts: 2363
Joined: Mon Apr 02, 2007 6:52 pm
Location: Braunschweig, Germany

Re: Achievable absolute accuracy of DAVID

Postby Khalid Khattak » Wed Apr 21, 2010 7:06 pm

WalterMo wrote:Khalid Khattak, I think you meant Gunter.
But let me answer, perhaps I am right. :wink:
The original diagonal length is, assuming the square was precisely made: (root of 2) x 299.4 = 423.416

Gunter, when we compare the measurements of the square item done by MeshLab and Rapidform explorer the difference is only 0.255%. For a first test it should be sufficient.

Walter


Sorry Walter for the confusion:)..

Now if the Original was : 423.416
The Meshlab calculation: 423.183 (0.233 from Original)
The RapidForm calculation: 424.264 (-0.848 from Original)

In the above case Meshlab calculation somewhat matches with the original one. Can someone tell me what the accuracy we can get from some other scanners? Is this accuracy enough for reverse engineering a part?

I am really watching this post closely.
PC: Dell Inspiron i3, 2.27GHz,4GB RAM,64 bit windows 7
CAMERA:
David CCD Mono USB
LASER:
David 5mW Green line laser
PROJECTOR:
3D LED Mini Projector 250lumens, Native 1024x600
[b]http://free3dscans.blogspot.com/
Khalid Khattak
 
Posts: 942
Joined: Thu Apr 23, 2009 1:30 pm
Location: Pakistan

Re: Achievable absolute accuracy of DAVID

Postby WalterMo » Wed Apr 21, 2010 8:23 pm

Now a scan of a calliper rule which shall fill the gap between the numberplate and the coin.

This time it was easier to catch the right marker points for the measurement with MeshLab because there were no edges which had to be searched.
The scan was made with the Pro9001 at 1200 x 1600 pixels. To get a better resolution the camera was rotated by 90°. Of course the object was also standing upright. Again „old“ calibration sheets were used. Scale length was 60mm. Scanned with the background panels.
The deviation from the real length was only 0.04%. :D

Walter
Attachments
Schieblehre mit MeshLab Marker1.jpg
Calliper rule with texture
User avatar
WalterMo
Moderator
 
Posts: 2363
Joined: Mon Apr 02, 2007 6:52 pm
Location: Braunschweig, Germany

Re: Achievable absolute accuracy of DAVID

Postby milz » Sat Apr 24, 2010 10:20 am

hello

Can this be a problem from the camera lens ?
If you measuring 1 cm in the center of the calliper rule
and at the border of the scan is there a difference ?

mfg milz
Xeon x5650@4ghz
Cam : USB2 CCD 1024x768 / USB3 1280x900
Laser : LC532-5-3-F(16x65) Focusable Green
Projector Acer X110P / Acer K132
dControl one David control box + Turntable + automatic texture lights
mooooom bathroom!!!!!!
User avatar
milz
 
Posts: 520
Joined: Wed Jul 15, 2009 12:24 pm
Location: Germany Cologne

Re: Achievable absolute accuracy of DAVID

Postby WalterMo » Sat Apr 24, 2010 1:09 pm

Hi,
Due to the calibration process of DAVID the lens aberration will be corrected. You see the three sections of the calliper rule and each shows nearly the same value for 1 cm. But don't forget, because of the high magnification of the scan and the texture it is hard to hit exactly the right point with the mouse for measuring.

Walter
Attachments
SchieblehrePanorama.jpg
MeshLab measurement at three different sections of the calliper rule
User avatar
WalterMo
Moderator
 
Posts: 2363
Joined: Mon Apr 02, 2007 6:52 pm
Location: Braunschweig, Germany

Re: Achievable absolute accuracy of DAVID

Postby WalterMo » Mon Apr 26, 2010 2:36 pm

And now from the XY plane measurements to the third dimension of an object, the depth.

For this determination Sven developed a nice tool which automatically displays the distance of two planes. It's a program which fits perfect planes to two scanned (.obj) planes and determines their distance (in mm).
I have tested it with the model of a 7-ary circular staircase. After the scan of the whole object each stair has to be separated, cleaned from any unwanted elevations and stored separately.

The real distances of the stairs were measured with a digital calliper. These values are in brackets.
The results in mm:
From step n to the next higher (more in the foreground)
1 9.196 (9.15)
2 9.3 (9.22)
3 9.31 (9.24)
4 9.3 (9.25)
5 9.44 (9.28)
6 9.3 (9.21)

I think it is a very good agreement. And we cannot expect too much, the staircase was made of wood which stairs aren't perfectly parallel in height.

Walter
Attachments
Wendeltreppe Foto.jpg
Wooden staircase
Beispiel mit PlaneFitting.jpg
Example with Sven's program "PlaneFitting.exe"
User avatar
WalterMo
Moderator
 
Posts: 2363
Joined: Mon Apr 02, 2007 6:52 pm
Location: Braunschweig, Germany

Re: Achievable absolute accuracy of DAVID

Postby Khalid Khattak » Wed Apr 28, 2010 4:18 am

Thanks Walter for showing us the superb scanning results.. Now i have to improve my setup because we can get accuracy with David Laser Scanner... :)
Keep posting your endeavors..my new gearbox is under manufacturing..soon i will be with you in this thread..
Regards
PC: Dell Inspiron i3, 2.27GHz,4GB RAM,64 bit windows 7
CAMERA:
David CCD Mono USB
LASER:
David 5mW Green line laser
PROJECTOR:
3D LED Mini Projector 250lumens, Native 1024x600
[b]http://free3dscans.blogspot.com/
Khalid Khattak
 
Posts: 942
Joined: Thu Apr 23, 2009 1:30 pm
Location: Pakistan

Re: Achievable absolute accuracy of DAVID

Postby WalterMo » Wed Apr 28, 2010 8:59 pm

And now something very hard for DAVID: Scanning of a 8-ary circular staircase standing in the open window of my flaps corner.

Each stair having a height of 51mm. That means the staircase has a total hight from the base plate of 408mm. About one half of the staircase was below and the other above the 17cm wide window. The question was, at what position regarding the height can we expect the best accuracy?

Used hardware:
Monochrome CCD camera from the DAVID shop.
Green focusable laser.
New V3 calibration sheets (146mm scale length).

Like in my post above the real distances were measured with a digital calliper and the scan data were „analysed“ with Sven's PlaneFitting.exe. The results in mm (calliper values in brackets):

from the base plate to stair 1: 49.14 (50.61) = -2.9%
from stair 1 to 2: 50.49 (51.16) = -1.3%
from stair 2 to 3 : 51.11 (51.22) = -0.2%
from stair 3 to 4: 51.04 (51.04) = 0%
from stair 4 to 5: 51.35 (51.16) = +0.4%
from stair 5 to 6: 51.00 (50.97) = +0.06%
from stair 6 to 7: 50.98 (50.62) = -0.7%
from stair 7 to 8: 52.08 (51.10) = +1.9%

It's again, like at the first staircase measurement above, a very good result. And what we could expect became true: The highest accuracy is in the middle where the side walls are close nearby.

Walter
Attachments
Wendeltreppe Foto2.jpg
Flaps corner with a model of a circular staircase
User avatar
WalterMo
Moderator
 
Posts: 2363
Joined: Mon Apr 02, 2007 6:52 pm
Location: Braunschweig, Germany

Re: Achievable absolute accuracy of DAVID

Postby hunkatibor2 » Mon May 03, 2010 4:18 pm

Image
Image
Image

Real sizes RHINO

16.9 17.3
17.1 16.5
17.1 16.23
17.1 17.59
16.6 15.9
17.3 17.07

100.5 99.78

90 spontoon turning

16.87
16.36
16.12
17.23
16.52
17.11

99.78
Last edited by hunkatibor2 on Mon May 03, 2010 5:37 pm, edited 2 times in total.
Intel Xeon E5410 2.33GHZ 16GB RAM Windows 7 64bit David 4.5.0
2 Basler A622F camera 2 Tamron Asfirical AF 28-80mm lense
LG DLP Projector PF1500G Canon D40 Tokina 17-70mm 1:2.8-4,5
Humanti Calibration panel system and Alignmed&Two camera
hunkatibor2
 
Posts: 1408
Joined: Tue Mar 16, 2010 4:26 pm
Location: Hungary

Next

Return to Camera Calibration

Who is online

Users browsing this forum: No registered users and 2 guests

cron