High Throughput Vision-Based System - hardware/camera advise

[lll]

Hello all,


i need to build a High Throughput Vision-Based System for analizing images
in size of 35cm X 2.5cm (resolution of ~ 1024X100 )

the object that i need to process is a very thin plastic cylinder
with height of 15-35cm, width of 2-7mm (the camera take image for one side only of this object)

the cylinder is curved, its color is in the range of the blue,
i need to find defects on this object, the defects are spots with different
color, i.e yellow or white. each spot can be in size of 1-40mm (above 3mm along the height of the object)

i use the MS paint to draw example of this object.
please see the attached image,the spots are marked with arrows.


the rate of processing and making the decision must be at least 20 images per second.

that means that i have 50ms for :
- taking the image by external triger
- process it with the NI vision module
* measuring the height of the object
* calculate the average color of the object
* treshold the image,by color-range and find particles in the image.
the particles are damages/spots in the material (defects in the production process)
spots with different color that not match the treshold
* calculate the Y axis of the particles that found (if exist..) and
get a decision to output digital pulse to one of 3 knives depending
on the Y of the spots that found in this image proccesing

So all is about rate and High Throughput.
20 images per second!
i know that the algorithm of the processing i build take
major part in the speed of inspecting the object,
but my questions are on the hardware,
i think that the software issues are solved for now.

Questions :

1. what IMAQ board i need to use ?
i though to use the digital board PCI-1422
with digital camera.
2. if i need a digital board, PCI1422, what camera i need ?
it must be area scan,digital, color,with external triger option camera.

i use the camera advisor and found that the JAI CV M7+ is ok,
but i not found camera file for this camera, there is one for the JAI CV M7 model.
is it ok for the CV M7+ also ?
what do you think about this camera ?
link : http://www.jai.com/camera/products_show.asp/id=7&sprog=uk

3. is IEEE PCI board and camera can work with this rate
of processing 20 images per second ?

4. the CVS is not an option in this machine vision system,
i limited to PCI board only.

5. i apprecite any other configuration of board/camera that can
handle processing of 20 images per second i just described.



Thanks in advanced

Regards,

Moti M.
Software Engineer

[BruceAmmons]

I think either the 1422 or the firewire solution will work fine for you. I usually prefer firewire, but they both work well.

I would suggest calling Graftek to figure out what camera will work best for your application. They sell a wide variety of vision hardware, and are very good at figuring out what you need. Their phone number is 800-441-2118.

One important thing to figure out is if you need color or not. With colors that have high contrast, a grayscale camera might be fine. A color filter on the lens may help increase the contrast between the colors. This might make the software end of the project a little simpler.

Bruce

[lll]

Bruce

Thanks for the very fast response.
If i can do the image processing i describe with firewire Camera & PCI-8252
i choose the firewire option.

there is a difference in cost between the digital and the firewire.
but the high throughput is esential.
in the knowledge base of NI the firewire is mentioned as "not fast" camera/board.

i need something fast as this :
http://sine.ni.com/csol/cds/item/vw/p/id/457/nid/124400
(Creating a High Throughput Vision-Based System for Pencil Sorting)

Is firewire camera/board is OK for this kind of application ?


Thanks for your reply

Moti

Message Edited by lll on 03-31-2005 01:46 PM

[BruceAmmons]

If you set it up right, I think a firewire camera would do the job. You would need to only send back the narrow stripe that you need to analyze to make the frame rate possible. With that, I think even a color camera would be able to keep up. I know a color camera can do 640x480 at 30 fps on firewire. Doing 1280x240 at 30 fps would be the same amount of data, and your requirement is lower than that. Going to grayscale would make it even easier.

Bruce

[lll]

Bruce

Thank you very much for your help.

I decide to buy the PCI8252 board with Sony firewire camera 640X480
and to configure it to 1280X240 as you advise.

I didn't know that its possible to configure and use 680X480 as 1280X240,
i though its limited to 680X480 or 680X240 etc..


Regards,

Moti

[BruceAmmons]

It is not possible!! You can't increase the resolution of a camera!!

I was just using the number of pixels as a comparison. The number of pixels in a 640x480 image is the same as the number of pixels in a 1280x240 image, so it would take an equal amount of time to transmit the data from the camera to the computer.

To get a 1280x240 image, you will need to have a camera with that resolution as a minimum. I would probably select a camera with 1024x769 or 1280x1024 resolution. You will have the camera send back just the part of the image you need.

Like I said, talk to Graftek and they will help you pick out the right camera for the project.

Bruce

[lll]

Bruce,

yes, after submiting my previous reply i notice that its not make sence.
my final decision is :
PCI8252
Sony DFW-X700 1024X768 Color, IEEE , with Trigger option.

i'm looking now in the PDF file of this camera to check if i can program the camera
to capture a ROI because my application need to take image of narrow object. 1024X96


Thanks for your help.

Moti

Message Edited by lll on 04-01-2005 09:46 AM