Scalable Spatio-temporal Analysis on Distributed Camera Networks

Abstract

Technological advances and the low cost of cameras enable the deployment of large-scale camera networks in airports and metropolises. A well-known technique, called spatio-temporal analysis, enables detecting anomalies on the largescale camera networks by automatically inferring locations of occupants in realtime. However, spatio-temporal analysis requires a huge amount of system resources to analyze a large number of video streams from distributed cameras. In particular, state update becomes a bottleneck because of the computation and communication overhead of updating a large application state. In this paper we propose a system design and mechanisms for scalable spatio-temporal analysis on camera networks. We present a distributed system architecture including smart cameras and distributed worker nodes in the cloud to enable real-time spatio-temporal analysis on large-scale camera networks. Furthermore we develop selective update mechanisms to improve scalability of our system by reducing the communication cost for state update.