Cheap Thermal Camera?/Best Way to Detect Location of Human/Animal?

[Keavon]

Thank you Fergs for your intelegent answer. I actually asked this question on an other forum, and I have gotten two other responces that is worth looking into, with one looking very promising. This is the post.
The first one is this:
"How about using an IR thermometer mounted on a servo? You can always use a range finder to determine the distance to the object. One of those thermometers can be easily obtained for less than $5. You would just need to hack it to make it work in your project. 1k times cheaper though.

Just a thought."

That sounds like a great idea, but the second sounds even more promising, though I do not completely understand what he means.

"Scan a pyroelectric sensor (as found in a PID sensor) with a grating/Fresnel lens back and forth (or better, rotate it) - that way it cannot be fooled by a person standing still.
It will detect even minute temperature changes crossing it and may have to be desensitized a bit."

When doing research on a PID sensor, it appears I can get one for $10, plus they are made exactly for tracking humans in range!
This page says "PIR sensors allow you to sense motion, almost always used to detect whether a human has moved in or out of the sensors range."
My only question is, does this know where the target is, or only if it is in or out of the sensor's range?

I hope it gives a whole image to base it off of. I will be programming something to run the whole turret on an old Android phone so I can have the entire thing self-contained. I will use an Arduino setup to run the stepper motors from the phone app's control. I will also have a BlueTooth connection from an other phone with an app I write to use as a remote control for more complex functions like making the turret dance and turn off and stop shooting or mute.

Thanks for your ideas!
-Keavon

[tician]

Unfortunately, no. Most (all?) PIR sensors cannot be used like that. A quick digikey search gave ~30 results, all of which had the detection system embedded within the sensor so that the user has no direct access to any of the 2 to 4 sensing elements. (I knew there was a reason the whole PIR discussion seemed way too familiar - this entire thread should be enlightening).

Hacking a cheapo IR thermometer onto a two axis turret to scan for temperatures like a laser rangefinder was one of the first thoughts that crossed my mind upon reading this thread, but the slow update rate would mean very coarse sensing at very long scanning intervals (easily enough to miss potential targets).

I seem to remember someone somewhere attempting to build an IR camera using discrete IR photodiodes or thermopiles set up in an array, but that may just be confusion on my part (I know that someone has made very basic X-ray sensors by coating photodiodes with a scintillating material).




Now I am very tempted to test out some optical flow algorithms on the new bot we got in the lab (combining them with zbar for reading QR Code tags on various objects around the lab may make the bot a bit more interactive). The basic fixed color blob tracking examples it comes with are not very interesting.

[mblackwolf]

If the turret is static you could still perform motion tracking without the worry of trees. You would need a controller with enough memory though. What is typically done is track the motion based on some sort of optical flow or background subtraction. To do simple background subtraction you could store a database of images for the background to get an idea of the movement of the background. This means small motion in trees would become part of the background. You then compared new frames vs. the databse to determine large scale motion such as animal/human motion. You could also threshold the tracking that it wouldn't consider something as moving if the pixels that changed between frame matched the color of a certain number of surrounding pixels which would eliminate small scale motion but not something like a human who when stepping into the frame changes a large group of pixels.