How did People Pong come about?

We've been playing around with Computer Vision for some time now at Cogapp. It started with our experiments with Augmented Reality and multi-touch tables. These systems use various techniques for blob detection, one of which is the OpenCV library .

The Open Computer Vision library was originally developed by Intel, and open sourced under the BSD license, and used by Stanford and VW in their winning entry for the DARPA Grand Challenge. It's one of those technologies that's so cool, it appears to be using magic. You throw images at it, ask it to find the faces in the image, and it just does. The library can be trained to detect objects in images. The classic example of this is human faces (note, this is face *detection*, not recognition, although there is a ton of interesting research into this area too). If you throw a ton of images at it, saying "these are definitely all faces", and then you throw a ton of images at  it saying "these are definitely *not* faces", then somehow it learns what makes up a face. The same process can be applied to anything, e.g. cars, plates, girls, or spinning leeks. And eyes...which is where People Pong comes in. Every other Tuesday, the mighty Cogapp tech team amass in the meeting room, get some lunch in, and geek out for an hour. These tech-tuesdays take a variety of forms, one of which is a coding Dojo, where we plunge into the unknown and attempt to create something out of nothing.

This particular week we chose to create a game of Eye-Pong using openFrameworks. What is openFrameworks? It's a collection of cool and powerful C++ libraries (one being openCV), all bundled up in a nice easy to use format. Not used to compilers and low level languages? No worries, OF takes up most of the strain and offers the user something more like a scripting language. Think Flash, or Processing, but with a lot more grunt. I'd been tinkering with OF for a while, creating a face swapping app (that produced some truly ungodly results), so briefed the team, and within an hour we had a rudimentary People Pong up and running. The rest was just tinkering and a slap of paint. The results weren't perfect (the eye detection only really works with your face straight on, glasses don't work well, many faces in the crowd confuse it), but overall we we were happy with the outcome. We ran it at our dConstruct stand last week, and it seemed to go down well...