Processing variable gray image

[chilly_charly]

I'm surprized by the scores you get with Vision assistant :
On my PC (Athlon 2.6 GHz, 512 ko RAM, W2000K) I find :
- Conveyor belt : 160 ms/image
- Conveyor belt 2 : 275 ms/image

Quite far from your scores ! May be instead of trying to improve the algorithm, you should try to improve the PC !

[chilly_charly]

Hum, I made a "small" mistake. I made some confusion between scripts :

conveyor belt

[chilly_charly]

Hum, I made a "small" mistake. I made some confusion between scripts :
- conveyor belt         56 ms/image
- conveyor belt 2    160 ms/image

Any comment ?

Message Edited by chilly charly on 07-29-2005 04:38 AM

[altelec]

HI CC
I’m sorry ,I believe I overloaded your mind, with my overextend and boring 9 points post!

All images I sent you were 355x355 pixels and 640x480 pixels. I'm talking about whole inspection area.
Read again, please, my item 8
8] And YES you are right, I’m afraid too. Now, I enlarged an image to a final size of 680 x 745 pixels

within an inspection area of 1250x960 pixels.

Voila cést tout ce qu'il y a de mystere:  I’m scoring an image four to ten times bigger!!!

In my PC, scoring 355x355 pixels images: CBelt=50ms/image and CBelt2=143ms/image

Yet, I need to stop many time consuming process running on XP, to improve score. And I read some

comments about best real time performance on W2000 over XP.

That’s correct?

Can you do any work with my last posted high contrast images?

Best Regards.

[chilly_charly]

Of course, playing with better contrasted images is a pleasure. I have attached 2 new scripts.

Conveyor belt High contrast should be close to optimal for shape, and is slightly faster than the previous script. I used manual thresholding, but in case of variable light level, it could be necessary to use an automatic thresholding procedure, searching for a minimum in the grey level histogram, between the object and the background.

Conveyor belt High contrast High speed does the job in less than 2 ms/image !

Of course some compromise could be found between these two extremes solutions.

I also think that with such well contrasted images, the direct pattern matching could work without preprocessing.

Message Edited by chilly charly on 07-30-2005 03:52 AM