Automated visual analysis in large scale sensor networks

Abstract

Modern automated video analysis systems consist of large networks of heterogeneous sensors. These systems must extract, integrate and present relevant information from the sensors in real-time. This paper addresses some of the major challenges such systems face: efficient video processing for high-resolution sensors; data fusion across multiple modalities; robustness to changing environmental conditions and video processing errors; and intuitive user interfaces for visualization and analysis. The paper discusses enabling technologies to overcome these challenges and presents a case study of a wide area video analysis system deployed at a port in the state of Florida, USA. The components of the system are also detailed and justified using quantitative and qualitative results.