University Student Creates Priceless Object-tracking Algorithm that Learns From its Mistakes

Imagine a facial recognition system that learns from its mistakes to continually track objects more precisely. Talk about intelligent search!

The Predator, an object-tracking camera system, can do that and more. Created by a University of Surrey student, Predator zeroes in on an object within a video by filtering out the other visual "noise." It intelligently learns what the object looks like so that it can continue to track the object as it moves or rotates. If the object leaves the screen and returns, Predator remembers and continues to track it. 

The system continually learns more about the objects it tracks, enabling it to match a face to a photograph or even follow your fingers, as it does in the video. 

We can't wait to see where Predator -- and its creator -- goes. 

Comments

[Rick Bentley]

The fact that it learns about the object as it observes it is what makes this such a great idea. You may have noticed that his three-finger-pinch got dropped the first time when he rotated it about the longitudinal axis of the camera, but it got better at tracking the same gesture later.

I've seen people/face/movement tracking fail miserably in the lab, we have a high end Japanese PTZ camera with built in motion tracking that serves only to track the shadows from our ceiling fan into a corner and then never come back out. This sort of learning algorithm is what is needed to make this type of technology useful.

Kudos to the creators.