GigE Vision Multithreading Problem

[krispiekream]

Hi Eric,

thank you for taking a look at my code.

The reason I went with producer/consumer because it was proven that it is faster using my webcam.

But I will recode it again and see if I can do what you suggested.

[krispiekream]

The program you attached seemed to run correctly on my computer.  However, the images were all the 640 x 480 size.  I do agree with you that the larger image size may be affecting your application. 

[krispiekream]

The program you attached seemed to run correctly on my computer.  However, the images were all the 640 x 480 size.  I do agree with you that the larger image size may be affecting your application. 

[krispiekream]

Ok.

I think I have recoded my program and dropped the producer/consumer and this seems to make it worse.

Can you show me with a diagram on how to change my producer/consumer architecture to have two queues of images?

[George_P_Burdell]

Krispiekream-

It appears you don’t have any pipelining occurring in your code.  I believe Eric was suggesting using shift registers to maintain only one image in memory and do your processing on that while acquiring in parallel.  

Eric-

Can you speak more to this?  

[krispiekream]

Okay,

Here is my acquired image. (I print the reference image and then use the GigE camera to view it.)

Here is what my reference image look like..

I thought about doing that to make the image smaller but then this doesnt work on Vision Asssistant.

I will attach the script file for Vision assistant.

i need to draw cross hair between the 4 corners as the object is rotated clockwise or counterclock wise.

i can make it work, except i cant make it work in real time.

it is very slow to process the image.

[BruceAmmons]

I would probably search for something smaller like the T, then connect the dots to get the crosshair.  That will be much faster than trying to match the entire pattern.

Second option would be to search for two opposing T symbols as a strip.

The circles do not make good targets since they are thin lines.  The T is a great target - solid with high contrast.

Bruce

[krispiekream]

i am having trouble using the reference picture with the acquired picture. can someone help me.

must i use my acquire picture and try to create the reference picture from that?

or can i use the reference picture, then acquire images and match with that.

[krispiekream]

i was able to do this.

my reference image is  this

i couldnt use this image.

The image wont detect when i am using the image above.

[BruceAmmons]

Normally it works better to create your template from an actual image.  It can be difficult to create a simulated image that matches the real image well enough to work.

I would consider masking the circle portion of the template so that only the T symbols are used for the match.  That might improve things a little bit.

What kind of processing rates do you get for this?  How many frames per second can you process?

Like I said earlier, it may be even faster to just search for a single T symbol and connect the centers.

Bruce

[krispiekream]

I actually tried the single T symbol and i got the same results.

I am thinkig that maybe Labview searches for all 4 Ts and matches the one that fit the closest to the 1 T.

I will try removing the 3 Ts and left only 1 T left and see if this will work.

I'll keep you guys updated.