A novel methodology is proposed to select the on-line, near-optimal positions and orientations of a set of dynamic cameras, for a reconfigurable multi-camera active-vision system to capture the motion of a deformable object. The active-vision system accounts for the deformation of the object-of-interest by fusing tracked vertices on its surface with those triangulated from features detected in each camera’s view, in order to predict the shape of the object at subsequent demand instants. It then selects a system configuration that minimizes error in the recovered position of each of these features. The tangible benefits of using a reconfigurable system, particularly with translational cameras, versus systems with static cameras in a fixed configuration, are demonstrated through simulations and experiments in both obstacle-free and obstacle-laden environments.