To predict motion of the "nose", if you will, I first converted the video image to binary and applied a blob tracker to find the largest blob. I then plotted the centroid of that blob, which appears as the face's nose.
To predict the movement, I applied a Kalman filter using the centroid and estimated covariance matrices to account for error. The filter provided me with both an estimated centroid position and a prediction for the future centroid position.
The face moves quickly in the beginning of the clip and slows down near the end. The predicted centroid is noticeably clearer when the object's speed is decreased.
Link to this comment:
All Comments (0)