I have done some successful motion tracking with a pan/tilt camera connected to a PC, but only tracking objects of a specific colour, i.e. following a red ball.
I have also done movement detection using a fixed camera, detecting any type of object moving.
What I'd like to do, it to combine the two, i.e. movement detection, then tracking with a pan tilt camera of any target.
The main problem is keeping the target identified as the camera moves to track it, because as the camera moves, the software detects the entire picture changing.
Perhaps a way to do it, is to grab two frames at a high shutter speed and a close interval and attempt to re-map the background images - i.e. re-align the two pictures and then compare for the movement of the target. Because I know how far the camera has moved between frames, i can re-align the images...
Originally, I tried stopping the camera, monitoring to motion then moving the camera to zero in on the motion, then stopping again. This was to stop the motion detection software getting confused by the movement of the camera itself. The results were poor.
Maybe a better approach is to identify the target while the camera is stationary and then look for a similar pixel pattern in each frame to try and find the target again. Trouble is, targets that move toward or away from the camera would be difficult to track, as would targets that change shape of colour as they move.
Any ideas on doing this, or does it require a massive amount of processing to do this kind of thing?