Vision application


Thread Starter

Stephen Wright

I have a couple of applications that require some vision detection and measurement. I have had less than glowing results from my local distributors. Would anyone care to share their successes or disappointments about vision systems? Particularly, one application is checking the
orientation of a flat metal disk where one side has a different groove pattern on it than the other. Another application is the detection of
scratches on shiny flat metal disks.

Any advice or sources of solutions would be greatly appreciated.

Matt Britter

You should talk w/ your local Banner (sensors) Distributor. Banner has a product called "presentence plus" which is a low cost pixel based sensor. We have succesfully used it in label detection and orientation. The success of your application really is in the though put of your device then having the correct light source.

Matt Britter

Anthony Kerstens

I don't have much to share, but I'm interested
in hearing what comes out. I have a need for
detecting defects in sheet steel as it's being
uncoiled. A defect could appear to be a small
fleck of dirt, or a scratch, or even a small edge
wave or indentation. At present, we have an
operator with push button. They seem to catch
almost everything, but almost is not good enough!

Anthony Kerstens P.Eng.

Davis Gentry

This sounds like a non trivial problem in vision - you can get a rough estimate of the level of the system by looking at resolution required and contrast available. It sounds like a black and white system MAY work - but only if you can generate good contrast between the smooth surface of your disks and the patterns/scratches. Side lighting may help. But this sounds like you have a low contrast, high resolution requirement (how big is the cross section of the scratches?), and may be a pretty expensive problem to solve. You may try something using other than visible light - is there a coating on the metal? If so, do
the grooves/scratches penetrate the coating? If that is the case you may be able to look at using UV or other wavelength which will be absorbed by the coating - I've done that with good success before, but it is not cheap.

Davis Gentry
Delta Tau Data Systems

Johan Bengtsson

What is the problem?

Anyway, this might or might not help you:

For the second problem: how much have you tested different light sources? a bright source shining almost along the surface will probably make shadows of any scratch that will be more easy to detect than if the light is located near the camera. You will probably need to check the same object with light from at least two different directions (90 degrees apart) to find scratches of any orientation.

Even the second problem might be easier with another direction of the light source.

/Johan Bengtsson

P&L, the Academy of Automation
Box 252, S-281 23 H{ssleholm SWEDEN
Tel: +46 451 49 460, Fax: +46 451 89 833
E-mail: [email protected]
We have had some good success with the 'Checker' vision inspection camera system at our plant. We use them to detect embossing and compound
distribution in aluminium foil on metal lids. They are quite cheap and easy to set up once you know all the pitfalls.

I will give you our guy's email so you can contact him directly. He set up the system and has looked at a number of other systems also.
[email protected]

Mike Lynch
Dairy Containers Ltd
Hi Stephen,
I'm a vision integrator and I can give you some help on the feasibility of your applications.
First of all, you have to appreciate if the feature you want to analyse is easiliy extractable from the scene or from the object.
The success of a vision application is based for me on 60% on the quality of the lighting and the rough image. After, you have to evaluate the necessarily resolution of your camera, the type of conveyor... For simple applications, counting small objects, detecting defects on manufactured products, you have numbers of vision PLC (matsuhita,Allen Bradley, Keyence...) which are easily programmable with an optical pen on screen, with some processing tools and outputs. A non specialist of vision problem can solve a vision application. If you need high accuracy, or if your products are long or if you can't stop the product, you have to look to linear system.
Programming this kind of hardware needs C, C++ ....

If you give me more details on your applications, I can tell you what kind of solutions wiil be fitted to your problem.

J-F Portala
SoViLor company
[email protected]

Dale Merritt

Try contacting CCS at 931-526-7914. They have completed a wide variety of vision apps successfully, but since I do not work with vision personally, I cannot tell you the scope of their capability.

Curt Wuollet

Begging the list's pardon,
We do these sorts of applications with definite purpose machine vision. The knowledge is more important than the product, the first should be
straightforward with custom software and the second will require care with illumination and perhaps filtering. Off the shelf solutions do some
things really well, but, if you are having problems, a program and setup specifically for that purpose can be the best way to go. We offer this on a consulting basis. Perhaps, for less than the off the shelf stuff.

Curt Wuollet,
Heartland Engineering. 612 389 9554 612 389 4440 X 4145 or email.
Anthony -

Check out and see what Cognex has done with the
systems that Isys used to make - Cognex bought them out a few years ago. Isys had the best continuous sheet inspection system on the market at the time. Not cheap, though.

Davis Gentry
Delta Tau Data Systems

Jimmy Saldivias

Hi Anthony:
Depending on how big is your sheet, you may use one or several camera system manufactured by Banner. Their website
Their sensors are called pixel counting. Here I type what comes in their own words:

"The PresencePLUS sensor is an advanced inspection system that captures a 256-level gray-scale image of a defined area, converts the image to white and black pixels, and renders a PASS or FAIL judgement of the image by comparing the number of designated-color pixels to a user-programmed reference count."

This system cost U$ 1000. A ridicule price for a vision based system.

If you want the controller -whose only function is to program the sensor- and lights and mounting bracket you will have to pay U$ 700 more.

But, you can use the controller with any number of sensors.

Also, if you want a more sophisticated system you can contact National Instruments. They have a vision system 1400 which includes camera, PCI board for your PC and software. But you will have to program this system. It goes for 7000 something. Plus you will have to install your PC.

Finally you have Their system goes for U$ 5000. I think their system does not need the PC either.

Or you can check with Their system also goes for U$ 5000.

Jimmy Saldivias
[email protected]

David McGilvray

Check out AutomationX ( from M&R Automation. They have a number of vision objects that provide a variety of solutions including both 2-D and 3-D applications. The problem you describe sounds like a relatively simple 2-D problem. A 3-D problems solved was the correct identification of different items from a single bin (screws, bolts, etc) regardless of the orientation with which they were viewed. Very complex problem but tremendously efficient for robot assembly application.

David McGilvray

Matthew da Silva

We've got a killer app, so advanced not even our CEO knows. Well, maybe the CEO knows, but only because he happened to walk into a secret demonstration while looking for the men's room. Our vision system uses Fourier transforms (*yawn*) and has a good applicability in electronic component identification. Orientation of target objects has zero effect on accuracy. Let me know directly, if you want to know more.

Matthew, Yamatake, Tokyo

Jay L. Mackey


I am the visions system designer/programmer for my company. We are a systems integrator, and have done assembly lines, test stations and robotic workcells that have incorporated machine vision.

I wanted to confirm some of what has been said on this subject. Lighting is the largest component of this application. The perfect vision system will still fail if the lighting is poor or misapplied, and I believe that for this app, even the crummiest vision systems on the market could get the job done if the lighting is done right.

For recognizing a specific pattern on a small (?) disk, I believe neuro-based systems are better. I will back up what another respondent
recommended: Eyebot Vision system by SighTech ( 408-557-6773) can do both parts of this application well. It is neuro-based and excels with patterns, but it also does a good job with defects. I saw it demonstrated at the Vision Show, and it was able to consistently detect hair-line scratches on the labels of small lotion bottles, and pass all bottles without defects. If you have product passing by single-file for inspection, it would definitely work. Because it is neuro-based, it is both very robust and very easy to set-up. And the price is competitive with the Keyence/DVT type systems.

However, there are other issues that you do not discuss. How is the product going to be presented to the vision system? What is the process? Is it important to provide visual feedback of the inspection to a machine operator? Is this an add-on to an existing assembly process? Are there obstacles to using certain kinds of lighting, or is the design of the vision inspection station completely open. What is the ratio of the size of a scratch to the necessary field-of-view of the camera.

For about the same as a cheapo Keyence, Sharp, Omron, DVT, or the new low-end Cognex ($5K-7K), you could put a frame grabber in a computer and use NeuroCheck software. A PC/framegrabber/3rd-party software solution will usually give you more freedom to display information that might be critical for the machine operator. For example, DVT can provide lots of visual feedback on a screen, but only if you leave a computer connected to it full-time, and any negatives that apply to a PC-based system would then also apply to the DVT/computer system. Another bad thing about DVT is the slow update of the picture even at the 10 M-bit transfer of 10-baseT connection.
It doesn't affect inspection speed, but if you need quick feedback of the inspection, DVT won't satisfy. It also irritates the fire out of me when I set up a DVT. Most of the PC software systems allow customizing of what is viewed and what can be changed by the machine operator. Additionally, Visual Basic or C can be used to create a simple or complex front end for
additional customization of the machine interface.

Another solution is also from NeuroCheck. It is another "smart" camera solution that incorporates a WindowsNT computer into a CCD camera. The
NeuroCheck software is run in the camera. the back of the camera has connectors for keyboard, computer monitor, ethernet, mouse, and I/O, I

RDP is a distributor for SighTech and NeuroCheck:

Hope this helps,

Jay Mackey
Spectra Technologies, Inc.
DFW, Texas, (817) 540-4475
I just read your message and realize that it has been almost a year since you posted it. However, I may be able to help you. Cognex specializes in vision applications. For more information, call 508/877-COGNEX1. Buddy Creef Cognex Corporation 508/650-3305

Bruce T. Dobbie

Pattern recognition can be difficult but not impossible. It sounds like your first application requires pattern recognition. You may want to try PC Industries in IL to see if they can help there. As to your second problem, subtle autonomous defects and scratches are indeed difficult and require very good experience and knowledge in eliminating product noise (frequency) from the defect itself. I don't know how successful you may be trying to develop your own system as many top manufacturers and suppliers of vision systems themselves have had trouble. Going to a local dist. for parts to build you own unit may prove more costly in the end if you even get to a position of developing a unit that works. Our company has specialized in discrete defect detection for some time. In particular, we have developed and patented a system that works very well for scratch detection. Of course, sample evaluation is always a prerequisite to ensure the job can be done correctly and reliably. If you are still having trouble finding a solution to your second problem, we would be more than happy to look at it. Of course, you have to be made aware that scratch detection, especially very subtle scratches may cost some money. It really depends on how small the defect, how many sensors required to do the job, the right lighting, process speed and of course, the right type of electronic processing circuitry to do the job. Recently we finished some tests with the Institute for Paper Science and Technology in which we picked up a coating scratch in a paper product about 120 lbs/3000 sq. ft @ 4500 fpm. This scratch when measured was 0.001" wide. We certainly would be interested in your application if you are indeed pursuing a solution and of course, you have the money to invest in such a solution. We wish you the best in finding a solution to your problem. Sincerely, Bruce T. Dobbie President R.K.B. OPTO-ELECTRONICS, INC.