3D Shape Measurement using two cameras

[Hansa]

Hello Folks,

I need to measure the Dome shape of the certain manufacturing process i.e. 3D Shape measurement (curvature) of the Dome shape object using two cameras. My experience with the LabView and IMAQ is basic. And I am looking for guidance or any related example, so that I can understand 3D shape measurement technique and reach my goal. Any help will be greatly apprieciated. Thanks in advance.

Manu

[David S.]

Hi Manu -

Unfortunately, NI doesn't provide any custom toolkits or functions for calculating 3D object contours that I know of.  An application like this would probably require a stereoscopic acquisition and some image processing functions to detect edges and contours.  If you do not already have a copy of the Vision Development Module, I recommend that you give it a look when researching this application.

Hopefully some of our forum gurus can help you further from their own experience with this type of application!

[Hansa]

Hello David,

Thanks. In fact, I am working with Vision Development Module. And I would like to have one example of some thing similar, so that I can get some direction.

Thanks.

Manu

[David S.]

Hi Manu -

I was unable to find any examples of stereoscopic image acquisition or 3-D contouring on our website.  However, you can find examples of how to use the NI Vision functions using NI Example Finder:

1. Open LabVIEW or LabWindows/CVI and select Help»Find Examples.
2. When the Example Finder opens, browse by Task to Toolkits and Modules»Vision.

You might also consider employing the help of one of our NI Alliance members.  The NI Alliance is a list of engineering contractors that develop advanced applications (like this one) using NI hardware and software.

[oliver_sidla]

Hi Manu,

IMAQ Vision does noot provide you functions directly to do 3D evaulation of stereo images.
We do 3D processing on stereo images for many years and have come up with several solutions, two of which might
be usable for you (I don't know the details of the shape you need to measure, so my answer will probably nit
fit perfectly to your problem, but I should give you a hint at least)

*) depending on how many points of the shape you have to measure you could use a light-sectioning approach -
   project a laserline on the cone and form its shape in the image calculate its shape. This woirks for one crosssection only,
but could be repeated while rotating the cone or the camera to get a full 3D set of points. We have implemented this algorithm
in LabVIEW and it works fastt and accurate.

*) (I assume that your object does not have any texture on its surface which can be used for stereo matching): you could project a kind of random pattern on the cone and use a stereo algorithm to match both images. We have implemented a
  version of the KLT point tracker in LabVIEW which is designed to do exactly this. You would provide a regular raster of points (a couple of hundrets, I guess) and match them using the KLT algorithm on the other image. From the resulting disparities you would get a  cloud
 of 3D poiints through which you would have to fit a suface (which should not be too hared in LanbVIEW) or do some other
processing to get your actual shape measurement (but I think this is the easier part of the solution). We have also
 implemented more sophisticated algorithms for stereo but these would be an overkill for your problem, I guess.

Hope that I could help ypu

Oliver

[K_seeker]

Oliver,

That sounds interesting. Can you send me the weblink for the products you are talking about? Or your company name. Either way I would like to save the link for future use. I work for a systems integrator. I beleive I could use one of these tools some day soon.

Thanks

[Hansa]

Hello Oliver,

Thanks for your reply. The shape of the dome is very much similar to top quarter protion of the sphere (but its not a perfect sphere). And I need to measure these uncertainities of the spherical form during the dynamic forming process i.e. the spherical shape is formed from the flat sheet. The true measurement of the spherical form is important to study the behaviour of forming process. And I also don't have the flexibility of rotating the part. So in my view, the second option suggested by you "KLT point tracker " may work. I need to measure the highest point (height) and the curvature of the form. Then I will apply some mathematical equations to perform further analysis.

I had read about this random pattern generation on test part and then tracking them with two cameras, as they move during forming process in the z-direction. But I have no clue about this KLT point tracker algorithm.  Any direction, whether it is a library function or something I have to code myself. Like where I should start and how!!

I would really appriciate your help in this regard and again thanks a lot for your reply.

Manu

[oliver_sidla]

Hi K_Seeker, Manu,

thanks for your replies. You can find information about us on 'dib.joanneum.at', for possible application of our tools please
contact me using oliver.sidla@joanneum.at. We have VIs which implement the functions mentioned earlier and build customized
applications/tools from those.

Manu:
Depending on the accuracy that is needed, the light sectioning approach might be more precise - what is the dimension of the spherical
form and what measurement accuracy do you need?
You cannot rotate your object - would it be sufficient for you to observe only 1 or very few cross sections? You still could project a few
laser lines onto the shape and observe only those.
 
You can find an implemenation of the KLT algorithm, which is very well usable, on http://www.ces.clemson.edu/~stb/klt/. This
site has also very good references to the algorithm itself. so that you can find out how it works. You can also email me if you have
more questions.


Regards

Oliver