MACHINE VISION BEGINS

3689411988?profile=original

 

Since getting video into a computer is cheaper than ever, we've been
contemplating the idea of a cheap, turret mounted camera for position
tracking.  The sonar transducer is a huge source of weight & it doesn't
look like Marcy 1 will ever be stable enough to stay in sonar range.
The reduction in board size & transducer weight would pay off.

The optical mouse is proof that machine vision may replace all other
sensors.  They 1st appeared 25 years ago, requiring grid mousepads.
They barely worked, 10 years ago.  Today, they're taken for granted.

Machine vision would have to difference key out the background.  It
would have to measure X, Y, & Z from a single camera.  It would have to
be cheap enough to justify the means.

After playing around with different color keying algorithms, was pretty
confident position sensing could be done better with vision than sonar &
could be good enough to keep Marcy 1 in a box.  Ideally the camera would
be turret mounted, so that meant recycling a micro camera to build
confidence before investing any money.

3689411898?profile=original

This 6 year old laptop camera was never used. 

3689411942?profile=original

Really nice LCD, if only we could use it in a new laptop.

3689412010?profile=original

Just a standard USB cam.

3689412070?profile=original

Reinforced & soldered.

Unfortunately, these old cameras only do 4fps at 640x480.  320x240 in
complete darkness gets them up to 15fps.  The price goes up as frame
rate goes up.  Position sensing is going to be 1Hz, at most.

3689411947?profile=original

Finally caved in & got a Logitech C510 camera for only $40, locally.
Turns out the Logitech C510 actually goes up to 1280x1024 even though
it's only advertized at 1280x720.  No surprise why it was for sale &
discontinued.

In bright light, with minimum contrast, minimum brightness, 0
saturation, it'll actually hit 30fps at 1280x1024.  The reality is 15fps
is the most our simplest machine vision algorithm can process.

It spits out tons of bad frames on our desktop, but not on the laptop.
Lacking a way to disable auto exposure is a killer for machine vision.

The object of the game is to detect the center & size of an object.  The
only algorithm that seems to work, at a decent framerate, is luma keying
in a totally black room.  It's certainly good enough, since Marcy 1 can
only fly in total darkness, but developing in total darkness is hard &
we really want to fly other vehicles in daylight.

Flashing LED's would probably work, but Marcy 1 is constantly moving.

3689412031?profile=original

Don't turn it on.  Take it a apart.

3689412096?profile=original

The stand needs to go.

3689412168?profile=original

It definitely has the Auto Pilot look.

3689412202?profile=original

3689412181?profile=original

Focus is glued in place.

3689412120?profile=original

3689411988?profile=original

Put it on a pan/tilt platform to try to increase the field of view.  The
problem is it can't use difference keying, the servo movement is going to put a lot of noise

in the result.  A fisheye lens would be ideal.

   

Now we have video of the camera turret tracking the object, & how well
it tracks.  Marcy 1 got a 2nd LED to give the machine vision a bigger
target.  Even though the Logitech C510 can do 1280x1024, we found 640x480 to be

the most useful resolution with the turret. 

 

Discovered the brightness on the Logitech doesn't buy any more high end.  It
just clamps the white levels at lower values & reduces the size of the
JPG's.  If there was a way to lock the exposure, we'd be ecstatic.

 

 

 

 

E-mail me when people leave their comments –

You need to be a member of diydrones to add comments!

Join diydrones

Comments

  • If by chance you need better camera, I would recommend Logitech C920. You can find high quality casing for CS lens here: http://lukse.lt/uzrasai/2013-07-modifying-logitech-c920-to-for-cs-l...

    3701777120?profile=original

    Modifying Logitech C920 for CS lenses
    Logitech C920 is a wonderful webcam. It produces good picture quality and has hardware H.264 encoder. It can do real full HD streaming (1920×1080 p @…
  • T3
    Nice EEVblog reference.
  • Excellent point Dale.... I think we're on to something.

     

    Step 3 is to make it spin.

  • Looks like the camera (from laptop or otherwise) is the ground station, tracking / ranging camera not going on the aircraft.   That will be pretty neat.   Seems to track pretty well.

    After the wood,  I think step two is to etch a circuit board.

    Keep it up, let's see more!

  • I'm not sure how well your laptop is going to fit on Marcy 1.

     

    It seems like step 1 in any of Jack's projects is:

     

    1) Take device, tape to piece of wood.

  • cant see the video... it has content from the media and isn't allowed in my country...
  • You're going to have major problems with most webcams and generic camera modules, as most are based on CMOS imagers that only support an Electronic Rolling Shutter. This will introduce major artifacts.
    Also, be prepared to pack some real processing power for any real work. Even simpler correlation and tracking algos usually end up requiring tons of processing power or dedicated FPGA design to run at realtime...
This reply was deleted.