0
$\begingroup$

This may or may not be an image processing problem. I am developing a simple system that captures images of a part as they are being manufactured and sent out along the conveyor belt.

I've solved the alignment problems, and background subtraction problem that this application requires. Everything is working in my prototype setup where I manually place parts with and without defect under the camera in different lighting conditions, position and other realistic variables. My setup uses Logitech C920 web cam, the software is able to process each frame in less than 1 second on average.

The problem arises when I start simulating the parts on the moving conveyor belt, it is unable to capture frame fast enough resulting in motion blur.

How is the issue approached in this industry? If I switch to use a DSLR and use the correct f-stop (etc.) will that solve the blur problem? If so, wouldn't DSLR overheat since it has to capture so many frames for hours and hours and hours?

PS: speed of conveyor belt is fast but slow enough to easily track with human eyes. I don't have the information about the exact speed at this time.

$\endgroup$
2
$\begingroup$

Try reducing exposure time if your camera allows that. With the short exposure time the conveyor belt should be well-lit.

If that is not enough, a common solution is to use a flash which is synchronized to the camera. If a sync signal is not readily available from the camera and you can't trigger the camera with your own sync signal, you probably could find a pin on your prototype camera's circuit board that carries a signal you can sync the strobe light to.

Without synchronization, setting the strobe rate to about 2/3 of the frame rate may be a reliable solution that gives good frames separated by at most one bad frame. Whether that works depends on whether there is a rolling or a progressive shutter and on the exposure time. Other ratios closer to 1 might work better in case of a rolling shutter and/or a short exposure time.

About DSLR cameras, you should use a camera that uses an electronic shutter in the mode (still/video) you would use. Cameras do not normally overheat but a mechanical shutter is only specified for that many cycles.

Also line scanners are a possibility, but I have no idea if those are commonly used. Think of a flatbed scanner where instead of moving the scanner head the target is moving. Those never blur the image but require bright illumination along the line being imaged. That may be easier to arrange than illumination over a large area, but it's a more exotic solution than a normal area camera.

More of an anecdote: A synchronously rotating prism with parallel opposite sides, located in front of the camera, can approximately compensate for linear motion.

$\endgroup$
  • $\begingroup$ went to do some research on line-scan camera. Looks like a potential winner. $\endgroup$ – 40Plot Dec 4 '15 at 8:57
  • $\begingroup$ If you know the belt speed, can't you deconvolve the motion blur? I anticipate a similar problem for my application but will know the exact motion speed using a belt encoder or linear stage. $\endgroup$ – user391339 Oct 25 '17 at 1:37
  • $\begingroup$ Another alternative to stroboscopic illumination is to stop/move the belt incrementally for your exposure time. I'm not sure what the ramifications of this would be for stepper motor versus dc motor driven belts. My guess is that steppers would handle the stop and gonwell. $\endgroup$ – user391339 Oct 25 '17 at 1:41
  • $\begingroup$ Perfect deconvolution of motion blur is not possible if the movement is 2 pixels per frame or more. The rectangular point spread function of motion blur is a sinc function in the frequency domain. Sinc has zeros, and as perfect deconvolution is frequency domain division, you'd have to divide by zero, which is not possible. But certainly you could still improve the image by deconvolution. $\endgroup$ – Olli Niemitalo Oct 25 '17 at 5:46
1
$\begingroup$

You usually don't use webcams for machine vision applications, but an industrial-grade camera.

You need a shorter exposure time. You need to make sure that during exposure the part moves by less than one pixel or so.

Reducing exposure time will reduce the amount of light, which you will need to compensate by

  • increasing the aperture, as long as possible and as depth-of-field allows it,

  • or adding illumination, preferably LED lighting, be it continuous or strobed.

These pieces of equipment are designed for long-term usage, but they belong to another range of pricing.

$\endgroup$
  • $\begingroup$ So are cameras (DSLR or not) with electronic shutter "Industrial-grade" enough? Since "one of the main design differences is industrial cameras use electronic shutters, while consumer cameras use mechanical shutters that have a much shorter life span, and higher failure rate" - lumenera.com/blog/items-of-interest/… $\endgroup$ – 40Plot Dec 4 '15 at 10:11
  • 1
    $\begingroup$ Indeed, I wouldn't use a camera with moving parts, nor with a "human interface", of little use in automation. USB3 and GigabitEthernet cameras are available. google.be/… $\endgroup$ – Yves Daoust Dec 4 '15 at 10:27
  • $\begingroup$ The cheapest consumer cameras like webcams only have electronic shutters. It's more a requirement than an indication of applicability for industrial use. $\endgroup$ – Olli Niemitalo Dec 4 '15 at 10:34
  • $\begingroup$ Came across "Vision sensors" as an alternative to "Industrial camera". Are those better? $\endgroup$ – 40Plot Dec 7 '15 at 3:11
  • $\begingroup$ Mh, don't know which "those" designates. Vision sensors have lower capabilities than vision systems/smart cameras. I don't think you will find one (sensor) that delivers the image. It all depends on your requirements for processing. $\endgroup$ – Yves Daoust Dec 7 '15 at 8:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.