A local extended Kalman filter for visual tracking

This paper applies estimation theory to the problem of tracking deformable moving objects in an image sequence. We extend previous work to derive a sub-optimal second-order curve filtering strategy. The second-order model accounts naturally for the curve velocities, resulting in better curve predictions. The second-order curve dynamics are nonlinear, so an extended Kalman filtering approach is utilized to estimate the position and deformations of a curve as it evolves in the plane. Application to visual tracking is emphasized through experiments utilizing recorded imagery and providing objective comparisons to other tracking methods.