32x32 CMOS Image SEnsor

[dknguyen]

Does anyone know of a non-obsolete 32x32 CMOS Image sensor (or some other low-pixel image sensor < 100x100). Preferably a low res one that can do high frame rates

Like this one for instance that is 32x32 pixels at 5000FPS:
https://www.ortodoxism.ro/datasheets/MitsubishiElectricCorporation/mXtvxwx.pdf

But it's obsolete...and I don't know where to get them, unlike their sister chip whcih you can rip out of Gameboy cameras. BUt it's a lower framerate and much higher pixel count.

 

[kchriste]

What about a 18x18 CCD out of an optical mouse?
http://www.diylive.net/index.php/2007/07/20/diy-mouse-scanner/

 

[dknguyen]

lol too low res! (I think).

I ran into an IC like that. BUt I couldn't figure out whether you could put a lens on it to have it work at long range. I don't think you can, because I think it needs an IR LED (or something like that) to illuminate. I want need something that can produce an image that, at 6-bits per pixel), at least several images can be held in the 30K RAM of a dsPIC. THe smallest one I've found so far is 128x128 pixels but that is 16K which is it too much. At 32x32 it's just 1K which is more feasible.

 

[kchriste]

If this to check for movement/direction in your chopper, then it is the perfect solution if you can get it to work at longer ranges:

The chip that has this sensor analyzes the signal, and can tell if the picture is being moved in the X direction, or in the Y direction. This chip then sends out an X and Y serial data pin.

If you want a camera to spy on the ladies, then yes, you need more resolution.

 

[3v0]

Use a higher resolution CCD and ignore the extra bits? What am I missing?

 

[dknguyen]

It takes a lot of cycles to average the extra bits down to form a single bit and processing power is already at a premium. I have to average pixels in the the rows as they come in. And then after all the "downsampled" rows are in memory I have to average the rows between each other. It only saves the square root of the memory it would have saved if it came in downsampled already since I need to be able to keep all rows in memory prior to averaging rows.

At lower pixel resolutions, it's also more important that the framerate is faster so I don't miss a moving pixel (I think) and high pixel resoltion and high framerate don't go hand in hand.

 

[mneary]

You can take samples at any interval you want. For a 320 x 240 ccd (example) with a 20 uS line rate, you can sample every 22 uS. Ignoring rounding errors and sync width, this gives you a string of samples that look like 32 x 24 at a 220 uS scan line rate. Do the downsampling by defocusing the lens.

If you have access to individual pixels, then just take representative pixels; but still do the downsampling with the focus.

 

[dknguyen]

If you did that, don't you get mis-represented images on highly varied images? Because I'm trying to track brightness changes. If I skip every X pixels, then if an "object (like a brightness patch or something)" has moved between pixels, I lose it don't I? Brightness changes between adjacent pixels are lost because the adjacent pixels are lost?

Like if you had fewer pixels, wouldn't each pixel kind of "average" out on it's own what it saw? Whereas if I skipped pixels I would see exactly what that one pixel saw but lose everything else around it?

It's also a problem to do it the way you mentioned because the pixels in the image are not at the same point in time which is a problem if I'm trying to track brightness changes.

 

[mneary]

If the ccd isn't focused sharply, any given pixel will see the average of a larger area. I would be careful to sample the same pixel on every frame, to avoid aliasing issues.

You'll get your 32 x 24 pixels on each frame at exactly the same time you would have, if you had been trying to average in software.

Somehow, you have to avoid being "in focus". Maybe a diffuser?