Visual Surveillance Monitoring and Watching

Abstract

This paper describes the development of computational understanding for surveillance of moving objects and their interactions in real world situations. Understanding the activity of moving objects starts by tracking objects in an image sequence, but this is just the beginning. The objective of this work is to go further and form conceptual descriptions that capture the dynamic interactions of objects in a meaningful way. The computational approach uses results from the VIEWS project. The issues concerned with extending computational vision to address high-level vision are described in the context of a surveillance system. In this paper we describe two systems: a passive architecture based on “event reasoning” which is the identification of behavioural primitives, their selection and composition; and an active architecture based on “task-level control” which is the guidance of the system to comply with a given surveillance task.