In looking at data correlation, we focus on the localization of the cameras in the network. We present a method for doing automatic localization by using the dynamic scene information these networks are now able to capture and store. Since the cameras in the network may be wide-baseline and not see similar static features, we use the dynamic scene data and detect moving objects in the scene. In an intra-camera process, we correlate the moving objects and build their trajectories within each image plane. These trajectories become the spatio-temporal features we then use in an inter-camera step by correlating them between the cameras in order to determine localization.
Regarding data integrity, we present a method using dynamic data for detection attacks on the cameras in the network. By doing intra-camera correlations as well as inter-camera correlations of spatio-temporal features, we develop a reputation system that is robust to the dynamic environment being observed, yet can detect when attacks occur. Our method determines when a camera has been attacked and is presenting faulty data.
Finally, in addressing data privacy, we look at the social concerns surrounding camera networks in public places and how video data affects privacy. We present a study on what privacy expectations individuals have in a public place and different factors that influence these expectations. Additionally, we look at different technical measures and examine whether they can uphold these privacy expectations.
Above all our hope is that this dissertation will aid in making current camera networks and dynamic scene information more beneficial. Additionally, we hope to inspire others to explore how computer vision can aid in real applications and go beyond the single frame, incorporating multi-view and dynamic information.