Integrated tracking and control using condensation-based critical-point matching

Image matching via multiresolution critical-point hierarchies has been shown to be useful in feature point selection, real-time tracking, volume rendering, and image interpolation. Drawbacks of the method include computational complexity and a lack of constraints on rigid motion. In this paper we present a method by which robot end-effector velocities are tracked using the Condensation algorithm and critical-point image observations. By using a window-based approach, we immediately reduce complexity while imposing constraints on camera motion. We show that the critical-point observations are successful in estimating camera motion by evaluating the similarity of sample windows.