Interesting post this morning about Zdenek Kalal’s “Predator” learning camera algo. The system uses an adaptive visual recognition and learning algorithm to dynamically identity and track any object in view. As soon as it’s tasked with a new object, it immediately begins gathering visual information about the image – the longer it sees and image, the better it can track it, even with size and perspective changes.
Next step: Pair this up with some of the work that’s been done with the Kinect camera system, and you’ve got the beginnings of a very Minority Report style tracking system..
Predator demo video:
Also interesting: Whiteboard Pong.