Visual Closed-Loop Tracking with Area Stabilization

We present a technique that combines image segmentation with nonlinear control to achieve closed-loop visual tracking. Camera rotation and zoom are used as the controls. The approach uses an analysis of admissible curve evolution flows generated by changes in projective scaling and angular velocity. Lyapunov stability analysis is used to synthesize a control scheme that renders the closed-loop system input-to-state stable (ISS) with respect to unknown target motion. Our formulation avoids the need to constantly perform rigid body camera calibration and motion reconstruction by stating the control goal directly in terms of image data. The results confirm expected tracking performance for a simulated 3D environment containing a camera and moving target.