Abstract
This paper describes an approach to detect and track multiple feature points for robust estimation of background motion under camera motion. The main contribution is a robust multiple features tracking algorithm based on KLT tracker and FAST feature point detector. We evaluate the tracking quality for each point and discard some points which are not well-tracked then add new feature points instead of them to keep the number of feature points during tracking. The features are combined with the segmented parts on the image to estimate dominant planes and homographies, and then separated into static and moving ones regardless of the camera motion. The camera motion compensation shows very good accuracy on a number of test sequences.
