Friday, July 20, 2012

3-D vision beats the clock

Whether you are short or tall, skinny or overweight, at one time in your life you have most likely used a pedestrian crosswalk to help you cross a busy street.

Using a crosswalk is undoubtedly a much safer bet than simply choosing your own spot to cross the road and risk being hit by a moving vehicle. Nevertheless, using such crosswalks can be rather an intimidating affair.

That's especially true of new those new fangled crossings commonly seen in Florida that incorporate a countdown timer to help pedestrians know just how long they have got to cross the road.

While they undoubtedly put the spring back in the step of many pedestrians who can then visualize just how long they have got before a hoard of Fords start hurtling towards them, they don't do much for the blood pressure of disabled or elderly people who may not be able to increase their velocity to beat the countdown.

Now, thanks to the help of a 3-D vision system, engineers at Migma (Walpole, MA) have come up with an interesting solution to the problem that has already been tested out at certain crosswalks with great effectiveness.

The system itself makes use of a stereo vision-based infra-red camera that can detect pedestrians during the day and at night. The output from the camera is hooked up to a computer that runs pedestrian detection algorithms that extract 3-D features of the human figures on the crossing from any other images that are present.

Detecting the pedestrians by using such a computer-based vision system enables the timing of the lights on the crosswalk to be controlled by their presence, a much more sensible approach than giving them an ultimatum which they may not be able to meet.

But is this really such a terrific idea? After all, how would the system be able to accommodate the behavior of those rowdy ne'er-do-wells who were intent on slowing traffic to a standstill by continuously walking back and forth across the crossing, perhaps in a protest to government cutbacks?

I think I have an idea. Perhaps what the system needs is an intelligent neural network-based back end that can "learn" to identify the behavior of such folks and then alert the authorities to take the appropriate action if its spots them acting in an untoward fashion.

Mind you, by the time you add up the cost of doing that, such a system might be just too gosh-darned expensive to make it worth installing. Back to the drawing board!

No comments:

Post a Comment