Vision network based surveillance systems are more and more common in public places. Typically, a mixture of static and pan-tilt-zoom (PTZ) cameras is used. Modern systems task PTZ cameras as a consequence of particular events needing further investigation; anyhow, the configuration of the network can be considered fixed and determined at the moment of deployment. In this work, we deal with a problem that has not yet been widely addressed: how a network can automatically change its configuration to enhance the monitoring capabilities. In particular, we propose a novel network reconfiguration algorithm that, given a map of activities, configures the pan, tilt and zoom parameters of all the cameras in order to improve the detection. A spherical model to project all the activities in the monitored area with respect to the optical centre of each camera is introduced. Such a model leads to an optimization problem that can be solved by means of the expectation-maximization algorithm and whose solutions are the new pan, tilt and zoom values for each PTZ camera. Experimental results will be proposed with both synthetic and real data to show how the proposed algorithm can be applied to different cases.
