The first four posts in this series have dealt with building my Raspberry Pi pan tilt platform, adding joystick control, colour tracking and face tracking.
Face tracking sounds neat but it has performance problems when run on the Pi CPU. My idea is to augment face recognition with something faster - perhaps motion detection.
Below are the metrics for motion detection
|Resolution||B+ Framerate / 10 frames (ms)|
|64x48||950 (min area=64)|
|128x96||2000 (min area=400)|
|160x120||2400 (min area=2000)|
Motion detection comes with its own problems however as every movement of the pan tilt platform requires the motion history to be reset. The motion history then needs time to rebuild to reliably detect motion again. Night time tracking is a problem due to CCD noise in low light triggering motion detection. This is a work in progress...