Different Approaches on the Implementation of Implicit Polynomials in Visual Tracking
In this work, we are interested in tracking a free-form object whose boundary can be described by a planar algebraic curve. We use a unique decomposition of algebraic curves to obtain feature points for position and orientation tracking. Decomposition represents such curves as a unique some of products of (possibly) complex lines. The real intersection points of these lines are so called "related-points", which map to one another under affine transformations.
We propose to do fitting only for certain frames in an image sequence and fill in the missing ones using Kalman filtering technique. Kalman filter uses the "related points" extracted from the decomposition of implicit polynomials of target's boundary curves and measured position of target's centroid.