An adaptive zoom algorithm for tracking targets using pan-tilt-zoom cameras

We address the problem of configuring pan-tilt-zoom cameras to track a target maneuvering in three dimensions; in particular, we propose an adaptive zoom algorithm that minimizes target localization errors by adaptively changing the camera focal length. The target tracker is implemented using a Rao-Blackwellized particle filter; the camera focal length is adjusted so that the images of a given percentage of particles fall onto the camera image plane. The focal length adjustment is also modified by a confidence factor that reflects the accuracy of the target position estimate. We evaluate the performance of the adaptive zoom algorithm using Monte Carlo simulations. These simulations demonstrate that the adaptive zoom algorithm has a smaller average squared position estimate error than a comparable fixed zoom algorithm.