Open ethanjdiamond opened 10 years ago
Have you considered using accelerometer?
That seems obvious now that you say it. Will report results.
We are working on a similar problem where we are testing a combination of CoreMotion and simple feature matching.
With CoreMotion we can positively detect pitch and roll, but yaw is too unstable due to the imprecise magnetometer.
So we use the full 3D transform, but correct the yaw error, by comparing the current captured frame with a reference frame, and see if we should offset left/right to compensate for the drift. Basically a 1D OTA implementation: http://www.ece.cmu.edu/~ee899/project/deepak_mid.htm
Anyone have any ideas on how to go about implementing a frame stabilizing filter?
I've tried combining a transform filter with the motion detecting filter, but it seems to have too much noise. I've tried using harris corner detection and trying to determine from corner movement how much the frame has changed, but it's hard to match up corners frame by frame.
If anyone's implemented this and could help I'd be very grateful.