We present a stochastic state-space quality metric for use in controlling active camera networks aimed at 3D vision tasks such as surveillance, motion tracking, and 3D shape/appearance reconstruction. Specifically, the metric provides an estimate of the aggregate steady-state uncertainty of the 3D resolution of the objects of interest, as a function of camera parameters such as pan, tilt, and zoom. The use of stochastic state-space models for the quality metric results in the ability to model and accommodate virtually all traditional quality factors, such as visibility, field of view, occlusion, resolution, surface normals, image contrast, focus, and depth of field. In addition, the stochastic state-space approach naturally addresses camera networks that are aided by other sensing modalities. We begin by surveying the traditional quality factors.We then present our new quality metric, aided by some background in the relevant stochastic state-space models, and an evaluation strategy that scales the computation of our metric to allow its use in a real-time active camera network system. Finally we present some simulation results that illustrate the incorporation of some traditional quality factors, and the use of our metric and evaluation strategy for some simulated scenes containing multiple objects of interest.