[ale] OT: Video Motion Sensor

[esoteric at 3times25.net (Geoffrey)]

Jeff Hubbs wrote:
> Using video, you could conceivably create an algorithm that would 
> estimate speed, bearing, and collision likelihood of an oncoming car, 
> but you'd be working at the edge of the state of the art 
> programming-wise and I think you'd still be fairly inaccurate (try 
> detecting both a motorcycle and a car with the same method).  You'd be 
> far, far better off with a radar or lidar rig.

Not to mention dealing with shadows, extraneous movement that you don't 
want in the equation.  Makes you realize how advanced the human brain 
and eye coordination really is.  A single camera is only going to deal 
with, for the most part, 2 dimensions.  It's a lot easier to detect a 
vehicle traveling perpendicular to your direction, but consider one 
coming straight at you.  They change in the image is substantially less 
apparent.  Add another camera at a fair distance away from the other, 
and you've got better data to make the decisions, but very complicated 
calculations.

I'll stick with relying on my vision and brain.... :) (and my 
mother-in-law's comments when she's in the car..)

I have a stationary camera that runs motion watching my driveway.  On 
windy days and partly cloudy days I get a lot of false positives.  You 
can mask sections of the motion detection out, but I've yet to find the 
perfect solution.

-- 
Until later, Geoffrey