Face Tracking Experiments

Some experiments with face tracking for my R2. The idea is to have the droid look for people around itself, then "lock onto them" and follow them around with its gaze.
The software can be downloaded at sourceforge.net.... Currently, it's in the testing/ subdirectory.
The main challenges are:
1) face detection is CPU hungry (even using LBP cascades instead of Haar cascades), which makes it rather slow on the kind of hardware that won't suck the batteries dry in an instant. So we need some kind of reliable technique to speed up recognition.
2) R2's dome is angled by 18°, which means the camera gets more and more angled as the dome turns, which results in a rotation of the camera images that classifier cascades really don't like at all. We need to compensate for this rotation.
Code is written from scratch using OpenCV. The camera is mounted on a servo driven by an Arduino, which is stuck to my R2's dome base plate. Only in the very first part of the video the camera is mounted to my monitor. The Arduino gets its commands from the PC running the face detection code via serial communication.
There's very little sophistication involved yet; no real controller driving the servo, no Kalman filtering the face positions or anything like that.
Further work:
1) try to recognize facial gestures such as nodding / head shaking.
2) try to get the frame rate even higher -- maybe involve the GPU.
3) set up an Eigenface face recognizer.

[@iRFANtgo]

ha ha awesome :D