Adaptive distributed video surveillance system

Abstract

Video surveillance systems are a major multipurpose video data source for modern emergency management solutions. As their complexity and image quality grows, the requirements for data exchange channels, storage and processing units are becoming much more demanding.Simultaneously, a wide range of neural networks becomes a universal instrument for image analysis. Such algorithms require deployment of powerful HPC-class server equipment.In this article, we propose a new adaptive approach to decentralized environment for video data storage and event-driven video processing for modern emergency management solution.The proposed solutions have been tested on different arrays of video data collected from various sources. These sources include city video surveillance systems, supermarket video surveillance systems, etc.For each of the listed examples, a discrete dynamic model has been developed based on video processing and event management.Our research resulted in a new data collection and analysis approach based on distributed video processing, reconstruction of 3D scenes and tactical situations detection.