GigE Vision Multithreading Problem

[BruceAmmons]

What are the steps of your analysis?  You must have more complex steps than me, or you are using a really slow computer.

Maybe post a screenshot of the performance meter with the details for each step showing.

Bruce

[krispiekream]

Okay,

I have done some more trial and error and this is what i find.

you use the method call Pattern Matching and i use the method call Geometric pattern matching.

With Geometric pattern matching, i get this.

But when I use Pattern matching, i get this.

Now, Pattern Matching is better!!!!! alot better.

But because I am using a microscope. I need to be able to zoom in and out.

is there something i can add to my microscope to autoscale for me?

Geometric matching allow me detect the template with zoomed in or zoomed out images (scaling of the template).

Can I still manage to use pattern matching with the scaling? 

[BruceAmmons]

I don't see any easy way to do standard pattern matching with scaling.  You could create a series of templates that are slightly different sizes and search for each one, but that starts using up all your time.

Another method is to skip pattern matching all together.  With the right pattern, you could do your own version of pattern matching.  Searching every few lines for the pattern, then fine tuning.

If all you care about is rotation, you could use a pattern of vertical stripes.  Extracting two rows a fixed distance apart would give you identical patterns with a shift.  Convolution would help you measure the shift, and then you can calculate the rotation.

It really depends what you need to do.  Are you just trying to measure rotation, or do you need to locate the position as well?  This is the first time you have mentioned scaling.  How big is the range of zoom?  How big is the range of rotation?  Are there any other requiremenst that you haven't mentioned yet???

Bruce

[krispiekream]

I need to measure rotation so i can calculate and draw the overlay. I need the x,y center point to draw the circle and the two lines.

the range is 1x to 6X. i am not quite sure how much to zoom in and zoom out yet.

the rotation would be 360 degree.

those are all the requirements i have right now.

if you can come up with a solution. i will be grad to hear it and would love to have you take on this project.

thanks

[George_P_Burdell]

I would suggest picking a center point and using the radial tools to determine how much your object has rotated.  The Developer Zone Article linked below may be able to help you.

NI Developer Zone: Rotating Part Example
http://zone.ni.com/devzone/cda/epd/p/id/5538

[krispiekream]

Wow,

that example is almost exactly of what I needed.

Too bad I really dont understand it.

Is taking the 2 days online class worth it?

Are there books for Vision Acq in labview?

[BruceAmmons]

Okay, based on your requirements I would suggest the following:

Create a pattern that doesn't change when you scale it.  One example would be a large circle with pie shaped wedges.  No matter how much you scale it, the center pattern will remain the same.  If you need to measure full 360 degrees of rotation, the pattern needs to be unique so it doesn't repeat as you rotate.  Right now, your T pattern repeats every 90 degrees, so you can only measure 90 degrees of rotation.  I would probably make a pie with a 90 degree wedge and a 45 degree wedge.

Once you have generated the pattern, train standard pattern matching in the center of the pie.  Train it at the lowest magnification.

Here is a sample pattern I generated.  If you train the pattern matching template in the center, scaling of the image will not affect the pattern matching.  It is also unique when you rotate it - it won't match itself until you rotate 360 degrees.

Bruce

[krispiekream]

thats really a great great idea.

but the bad thing is...

in the center is my circle. a fixture to put my piece in.

the center is a circular diaphram.

the cross hair helps the operator place in exact 4 corners of the lines.

[BruceAmmons]

Can you post a picture of the actual fixture, fully zoomed in, and another picture fully zoomed out?  This might give us a much better idea of what we are working with.

Right now I am thinking a circular ring target around your hole that has a black and white pattern to determine rotation.  The first step would find the circle (and its center), then the second step would measure the rotation.

If you just added a wide ring around the edge of the hole with 90 degree sections alternating black and white, couldn't the eye connect the edges and locate the center without help from the computer?

Bruce

[krispiekream]

We have'nt fully developed a pattern that we are certainly happy with.

but this is what we have in mind.

1X

2X

and our center is something like this.

So ideally, we place the object in the center of the fixture above.

we use pattern mattching and ignore what is in the center.

now we are only facing with SCALING issue when zooming in  and out