Machine Vision

cancel
Showing results for 
Search instead for 
Did you mean: 

Inspection speed

Solved!
Go to solution

Hello,

I am trying to measure 80 hole diameters in 1 second (One frame at the time).  I know I can get a camera that I can trigger that fast, but what about the inspection time.  On my present Laptop, a Toshiba with a Centrino 1.76Mhz processor, just the "Match Pattern" function takes 40mSec (using the Benchmark function). 

My question is: How do I calculate what processor I need to reduce just the "Match Pattern" function to 10mSec?

Will a "core i3" or core i5" do the job?

Thanks,

gertSC

 

 

0 Kudos
Message 1 of 8
(4,594 Views)
  1. In my opinion, it does not seem right to measure a hole diameter using pattern matching. Can you post an image how your frame looks like?
  2. Which tool do you use for designing your application? LabVIEW? Vision Builder? In most cases you might take advantage of multiple processing cores.

View my profile on LinkedIn
0 Kudos
Message 2 of 8
(4,583 Views)

Hi,

Actually I am not using "Pattern Matching", but "Geometric Matching" to locate the hole (it is not always in the same place), then "Coordinate System" and then "Circular Edge" to get the diameter. 

 

I am using Vision Builder 2010

 

thanks,

gertSC

 

 

Download All
0 Kudos
Message 3 of 8
(4,575 Views)

For clarification:  Are you capturing 80 images per second, and locating and measuring one hole per image?  Is it different parts, or is the same part moving around?

 

Pattern Matching may be faster than Geometric Matching since Geometric Matching has a certain amount of preprocessing that must be done to find the edges.

 

If the hole is pretty close to the same location each time, you could use other tools to locate the hole.  Passing vertical and horizontal lines through the expected center of the hole will give two strong edges on each line.  The actual center axis should be halfway between each pair of edges.  This will get you close enough to center to measure the diameter.  If the location changes more than 1/2 the diameter, using several lines space 1/2 diameter apart could work.  There is a little more processing figuring out which line has the best edges, but it is essentially the same algorithm.  This might get complex enough to require LabVIEW, though.  You might be able to program it as a LV block in VBAI.

 

Bruce

Bruce Ammons
Ammons Engineering
0 Kudos
Message 4 of 8
(4,556 Views)

Thanks for the response,

Yes, I am capturing 80 images per second on a curved device device that is spinning. 

What I really want to know is what improvement in inspection speed can I expect going from a Centrino 1.76Mhz processor to a "core i3" or core i5" Processor.

 

Thanks gertSC

 

 

0 Kudos
Message 5 of 8
(4,541 Views)

The improvement is impossible to predict.  The difference in processor speeds is one factor, and multiple cores is another.  However, doubling your clock speed doesn't always mean you double your inspection speed.  Likewise, two processors aren't going to double your speed.  It all depends how the code for VBAI is written to take advantage of these.  Since VBAI is essentially a linear process, I wouldn't expect dual processors to do much for you.

 

If you port your inspection to LabVIEW, you can make some improvements.  You could make the acquisition parallel to the processing and even queue up a few images if the system lags momentarily.

 

I think our suggestions for speeding up your analysis would do you more good.  Pattern matching is a slow process, and finding alternative solutions can greatly speed up your analysis.

 

Bruce

Bruce Ammons
Ammons Engineering
0 Kudos
Message 6 of 8
(4,539 Views)

I have a Intel Xeon 2.67GHz 4 core computer and the following algorithms took the following amount of time in VBAI:

1.6ms for Detect Objects - I would recommend this one since your object is easy to detect and this step will return the coordinates for the center of the circle (The find edges suggestion of Bruce would be even faster, but would require additional processing to find out exactly where the center is).

3.2ms fopr Pattern Matching - using default

5ms for Geometric Pattern Matching - using defaults

 

Hope this helps,

Brad

0 Kudos
Message 7 of 8
(4,535 Views)
Solution
Accepted by topic author gertSC

Thank You, I will change the inspection method together with a faster processor, that should solve it.

Thanks again,

gertSC

 

0 Kudos
Message 8 of 8
(4,532 Views)