IR human body recognition

Status
Not open for further replies.

cosinus

New Member
Hi all!
I am a student and as a final project me and my partner are trying to construct a robot which orients in space using ultrasonic sensor and is able to recognize a human with infrared detector.
The question is - where to look for info on this (IR) type of detection? Maybe somebody here dealt with this kind of problem? Which parts to use, how to approach, etc. Any help is appreciated.

.x-posted on All About Circuits forum.
 
What's your working model of the human IR characteristic? Got any way to tell a human from a chinp? On a warm day, a human may be cooler than the surroundings - how do you deal with that?
 
PIR sensors don't look at whether it's a hot or cold body they just look at a large rate of change in the temperature of the surroundings. Imaging looking at an image though a thermal imaging camera on a hot day and then someone walks past. The surroundings may be hotter than the human but you'll see a sudden change in the picture and this is how PIRs work.
 
Actually IIRC the PIR sensor not only senses a change in the temp, but due to its typical lobed pattern it will sense a change anywhere in the viewing area. That is to say that a man starting on one side of the viewing field moving to the other will still trip the sensor, even though the average value of the infrared emissions in the viewing area remains the same.

It cannot detect a nonmoving object. Typically the time constant seems to be set so that changes over more than tens of seconds are filtered out. You can play a fun game where if you walk up to an IR sensor really, REALLY slowly it won't detect you.

PIR seem ill-suited for mobile applications such as robots. Any objects in the room which are hotter or colder than the surroundings may trip the sensor because the sensor is moving, not the object.
Perhaps you could turn on the PIR only when the robot comes to a stop and allows a moment for the PIR to adjust to the surroundings and become sensitive to IR changes. However it could park itself right by your leg and won't see you if you don't move.

There's IR sensors like the noncontact thermometer, gets a temp from one target point, but it won't detect people well. A person wearing clothing may only register a few deg above ambient on their clothing.

Otherwise I mean there's IR cameras like FLIR, quite expensive in themselves, and then you need some heavy image recognition software and powerful hardware.
 
Case in point - I've had PIR activated light switches turn themselves off when I've been in front of a computer long enough to blend into the switch's time constant.


- nice picture of the bare element. If you were to put a normal fresnel lens instead of the 'segmented' one that is in most PIR motion detectors, and add in a temperature reference, you'd have a standard IR therometer.

Long-wave infrared (i.e. heat) cameras are made up of arrays of the same IR sensitive material stuck onto a chunk of silicon, and the whole assembly is called a Focal Plane Array. They are mind-numbingly expensive, and when you get down to the details, there isn't too much different between this and normal machine vision.
 
Thank you for the replies!
I don't have anything more to say yet, but I'll keep updating the thread when there's more info.
hjames, you gave me hope! .)
 

Intel has recently released an open-source computer vision library--more information at http://www.intel.com/technology/computing/opencv/. There's a slightly friendlier wiki at http://opencvlibrary.sourceforge.net/. Servo magazine is currently running a tutorial series on this library, which is how I found out about it.

There's a thread about a human/pet tracking project at **broken link removed**.

Haven't used it but it looks really fun. It appears to make face recognition a lot easier than I would have thought reasonable.

Of course, that doesn't help with the IR part of it. Does it have to be a heat camera image you display, or does it just have to work in the dark? If the latter, then maybe you could get away with using a bunch of IR LEDs to illuminate the scene. A webcam should be able to see the image (they usually aren't limited to visible light). Or just get one of those IR-ready surveillance webcams--the kind with a ring of IR LEDs around the lens. The images don't look nearly as cool as from a heat camera but this might satisfy the IR requirement.

Just some thoughts,


Torben
 
Status
Not open for further replies.
Cookies are required to use this site. You must accept them to continue using the site. Learn more…