A Novel Method for Scene Modeling to Detect Unusual Activity

Abstract

Abstract. Automated video surveillance is crucial for the security of various sites including airports, train stations, military bases, and many other public facilities. A modern surveillance system is expected to not only perform basic object detection and tracking, but also to interpret object behaviors. This higher level interpretation can have several applications including abnormal behavior detection, analysis of traffic trends, and improving object detection and tracking. In this paper we focus on the problem of interpreting the output of the object detection and tracking module in order to gather knowledge about the scene. This knowledge is used to build a scene model which can be used to detect abnormal motion patterns and to enhance the surveillance performance by improving object detection. We present two novel and complementing models here: first model that is suitable for modeling single object motion, and real-time applications and second model that is useful for learning relationship between concurrently moving object pairs in the scene.