An adaptive normalcy decision for video surveillance security system

Abstract

This paper proposes a robust and adaptive method of normalcy decision in video surveillance security system. It is a challenging problem since no technology can provide satisfactory performance up to now in the state of art of video security surveillance system, especially in automatic normalcy decision. The implementation of the intelligent security using the framework of video surveillance needs people detection, tracking, and normalcy decision. This paper focuses on the autonomous normalcy decision after people detection and tracking have been carried out. Computer vision techniques are used to analyze video streams of multiple people movements acquired from multiple video cameras developed by GE GRC's (GE Global Research Center). The extracted trajectories from people movements are used to construct trajectory ontology, and to monitor and produce alarm based on trajectory ontology. The system works incremental way by adapting itself toward both varying environments, and provides interactive interface with changing security strategy. Experiments were performed on outdoor video sequences with different illumination conditions in GRC's laboratory environment, and we achieved very encouraging results.