Design and Evaluation of a Notification System for Alarm Management in Distributed Vision Networks

Latest research results in the areas of computer vision and embedded systems have made way for smart cameras. These cameras use a built-in computing unit to analyse images directly at the CCD sensor. Smart cameras can cooperate in order to analyse scenes from different perspectives and track objects autonomously. In large smart camera systems (also known as distributed vision networks), distributed algorithms are needed that allow for cooperation and self-organization among cameras.This paper focuses on alarm management in distributed vision networks, in which security staff is equipped with mobile terminals. These terminals and cameras are connected via an ad hoc network. We present a robust and lightweight algorithm that allows to notify guards quickly in case serious incidents have happened. This algorithm has been evaluated by simulation experiments and in a modestly sized real world testbed. Results show that the algorithm takes reasonable time to terminate (10s) in a network of 100 cameras.