Robot picking a person from other objects

Question

I need some ideas of sensors to use so my tank can drive towards humans and seek them out.

Distinguishing them from the background.

Assume the human is standing still / quitely

Any ideas?

• looking to use an Arduino as the controller - though if there are better options please let me know.

Answer 1

A smart sensor might help -- take a look at the CMUcam.

Answer 2

Could you use an array of passive infrared sensors?

Ladyada's got a bunch of stuff on interfacing with them, including Arduinos.

I think the Arduino only has 6 analog inputs, but you could multiplex more sensors in and out using FETs if you need them.

Answer 3

I don't believe the arduino has the horsepower to do this. It's a very difficult task you propose.

Answer 4

OpenCV as zklapow says will allow you to use Haar Training to train the camera/controller to recognise faces or other shapes (this is how face recognition is done on my camera I think) but I'm not sure what resources there are available for Arduino. There's what looks like a fairly detailed discussion here using OpenCV. There's an interesting looking page here which is talking about haartraining and robots (recognising hands to be specific).

Answer 5

An infrared camera would be perfect.

Answer 6

What kind of platform are you using to run the robot. If you are running a like an ARM processor (with linux) than you can use a camera and OpenCV, but if you are using an Arduino or somethings along those lines that wont work. An arduino will be tough and definitely wont run OpenCV.

Answer 7

You can hook up a camera that sends serial data to the Arduino (available from Sparkfun, I think they are normally used for cell phones, it looks similar to the CMUcam solution) then send data to your computer via usb, or you can use a wireless module if your feeling swanky. Then you could download a facial tracking program, that would process the images from the Arduino. Facial tracking is a well established method of real time video processing, it should be relatively easy to find one. That's the only way I can think of that may get around the Arduino's lack of horsepower, you will need to be hooked up to a proper computer tho, if you want to process realtime video. It would be more expense and effort compared to using NIR sensors, but the results may be more suited for your application.

Answer 8

One strong predictor of Human-versus-Object is that the human body's temperature often lies in a specific range that is different from most objects' temperatures. Exploiting this fact, you can try one of the following two ideas, which I think should have good accuracy and straightforward implementation:

• Option 1: Use an IR temperature sensor (small and inexpensive) like Texas Instruments TMP006 or alternatively Melexis MLX90614
• Option 2: Use a prepackaged solution like Panasonic Grid-eye, which is an infrared array sensor that provides an 8X8 pixel representation of temperatures sensed in its view. The part is inexpensive (and available from distributors like Digikey) and, because the dataset isn't very intensive, you can even interface the sensor with a microcontroller board like an Arduino.

Using either of these sensors, and comparing the sensed data against a pre-defined window (for human body temperatures), you should be able to tell fairly accurately whether an entity is an object or a human.