View Full Version : motion tracking

Grand Robot Master
05-03-2009, 02:47 PM
is there any sensor out there that can track just motion, preferably not color, from a decent distance?
I'm looking for a sensor that can track a person. Would body heat be good for this?:confused:

05-03-2009, 05:29 PM
You and everyone else on planet earth...

Body heat can be tracked with a thermopile - unless there's something else with the same heat. Like other people.

Motion can be detected with a camera (change in pixels) or with microwave or capacitance motion detectors. Passive motion detectors like you use on alarm systems are cheap. When they see motion, they turn on. When they don't, they turn off. (or vice-versa depending upon how you wire 'em.)

05-03-2009, 07:36 PM
People will need more detail to help. What is a "decent distance" ? Track just motion or "how much" motion. Does direction matter? Etc. If you give more details then you will probably get more helpful advice.

Grand Robot Master
05-09-2009, 11:09 PM
answering you detail reply, perhaps about 10 feet or more, a rather sebnsitive motion enough to track a human.

05-09-2009, 11:12 PM
"Tracking a human" will also "track a deer", or "track a butterfly" if it's a few inches from the camera. I think Matt was trying to figure out what you wanted to accomplish, like the end-game.

05-10-2009, 01:24 AM
I think the real problem is an understanding of robot vision...or well...more the limitations of robotic sensors.

Vision: Using a camera is not like using your eye. What would you tell computer/robot to look for? Say you look for motion...that would be a change from one frame to another...You can tell where in the robots field of vision the motion took place, but that doesn't tell you the distance to the change, or the size of the change. Couple that with the fact that if the robot itself moves you will either detect nothing but motion, or you will have to temporarily stop looking for motion, and be "blind."

Looking for a color is just as bad. Unless the lighting conditions are really good, then you have too options. You can either have the robot look for a wide range of colors in which case it will be easily fooled, or you can have it look for a narrow range of colors, in which case it will not be very reliable. Slight changes in lighting due to things like time of day, or a door being opened or closed will make the bot "blind"

You can do edge detection...That is looking for changes in color from pixel to pixel and trying to line them up. So basically you would teach the robot a shape. Well...as things move the relative shape they are changes also. If you look strait down at your garbage can, it looks like a circle. if you look at it from the side it would be a oval. or a rectangle...

so vision probably won't work out very well...atleast by itself.

You could use a long range IR sensor, sonar, or even Lidar...but that would really just tell your robot how far away things are.

RFID would require you be very close...as in centimeters...and I don't think it would be directional.

I guess your best bet would be to use a number of different sensors coupled together. IR would tell you how far something is...and then edge detection could be used to figure out how big it is. If you have a strange colored shirt...you could use color detection...provided you were in a room that controlled for lighting...

I hope this helps.


Grand Robot Master
05-13-2009, 07:57 PM
how about PIR sensors?

05-13-2009, 09:28 PM
They do a moderately good job of passive infrared.

See: " think Matt was trying to figure out what you wanted to accomplish, like the end-game. "

05-13-2009, 09:44 PM
Note that all PIR are not created equal. The $10 models from parallax and such, require about 2-3 seconds to update (they are really intended to be motion detectors). The $65 E442-3 that acroname sells would probably be a better match for "tracking" motion. Regardless, none of these sensors will say definitively "here is a person", just "I passed an IR source or an IR source passed me"


05-14-2009, 10:03 AM
Just to add to DB's vision discussion.

You could use a algorithm like SIFT *Scale Invariant Feature Transform* on the object your trying to track. SIFT itself is a fairly complicated algorithm to write yourself, but luckily many people have written it for many different languages. I'm currently implementing a c++ version on ubuntu for alfred.

However for sift you need a template image of everything you want to track. It's also a little bit at finding an object even when the object is rotated or there is are changes in lighting conditions.

another technique would be to do back ground subtraction. Basically you take two images and compare them every second or so, if there is a difference between them you have motion. This can be a bit more robust against changes in lighting conditions and doors opening and closing.

Grand Robot Master
05-24-2009, 03:31 PM
here's an idea: what about taking a bunch of IR or similar sensors, and instead of measuring each sensor's value, could you measure the change in the "tripping" of these sensors?

05-24-2009, 04:37 PM
I think you'll need to elaborate more on that.

Have you researched and understand fully how "IR and similar sensors" function?

Grand Robot Master
05-24-2009, 06:25 PM
ok: let's say that there are three sensors: A, B and C, arranged with A on the left, B in the center, and C on the right. if an object is detected by B, then instead of reacting to the activation of B it would react to whether A or C is tripped and respond my moving in that direction

05-24-2009, 06:31 PM
Sure, you could do that.

Grand Robot Master
05-24-2009, 06:47 PM
also I was looking at some sensors and found the Sharp GP2Y0A700K (on acroname robotics) i thought it may look ideal. any input?

05-24-2009, 09:31 PM
Ideal for what?

05-25-2009, 01:04 AM
Ideal for what?


I believe the original spec was: ""perhaps about 10 feet or more, a rather sebnsitive motion enough to track a human.""

So @10 feet, yeah, that sensor is the only sharp IR that is gonna reach that far -- but note the current consumption a whopping 0.3A.. thats right, AMPS. Also.. how is 3 IR sensors gonna be able to track "rather sensitive motion"... The beam on these IR's might as well be a lazer for what you are proposing, the granularity is gonna be huge.

For that matter, why not a LIDAR, meets the 10 ft, should be able to track a person with no problems.. might hurt the bank account a bit...


Grand Robot Master
05-25-2009, 09:12 PM
what do you mean by granularity?
and about lidar sensors, where can i get one, are they easy to work with, how much current do they draw, and how bad will they shrink my wallet?

05-25-2009, 09:26 PM
$8000 to $10000 is kinda the minimal entry point for a decent one.

They're pretty easy to work with as long as you're a decent programmer.

granularity = try typing it into google and reading the definition

Current draw depends on the device. Type it into google and read the spec sheet.

Grand Robot Master
05-25-2009, 09:33 PM
yeah, this project is that paintball or airsoft sentry gun. it should be fairly simple, run a voltage at about 7-12 volts (i could push this though) weight is barely a factor, but it should be fairly easy to carry.

Grand Robot Master
05-25-2009, 09:39 PM
sorry but
A,my resources cannot afford that
B, im not a decent programmer
C i can't seem to find information in this sensor
but, in short, this sensor is rather inefficient for this project.

05-26-2009, 12:08 AM
So - it's a paintball turrent. What happens when a bush is blowing in view of the sensor? Bush = Person = Deer = fog = rock thrown at the sensor.

05-26-2009, 01:07 AM
If it was fairly simple? What you trying to accomplish is far from that. A $100 sensor is not going to accomplish what you want. No matter how many sensors you think you can hook up. At this point I would say your current project is dead. You need some really smart sensors and a very decent program at least to run this.

A wise man once said "You'll shoot your eye out" he comes around evey year on TBS.


Grand Robot Master
05-26-2009, 01:57 PM
i see.. all the one's ive seen have miniature computers on them......very complicated and expensive. i was trying to find out if there is a way to make one that is simpler and cheaper. probably wouldn't work as well but good for the money......i guess i cant.