Cooperative Visual Surveillance Network with Embedded Content Analysis Engine

摘要

Visual surveillance plays an important role in security systems of digital home and enterprise Regazzoni et al. (2001). Evolving from CCTV video surveillance, the IP camera surveillance system with Internet as the connection backbone is a trend in recent years. A typical IP camera surveillance system is shown in Fig. 1. IP camera systems have the advantages of easy setup and universal access ability; however, several issues in network transmission are introduced Foresti & Regazzoni (2001), which become more and more important when the number of camera grows. Since the surveillance systems share the same network with other applications and devices of digital home and enterprise, the congestion of network caused by transmission of large surveillance contents may degrade the service quality of the these applications, including the surveillance application itself. Besides, the control server can only afford the content storage from a limited number of video channels, which limits the system extension in camera number. Further evolving from IP camera systems, the maturity of visual content analysis technology makes it feasible to be integrated into the next-generation surveillance systems to achieve intelligent visual surveillance network Mozef et al. (2001) Hu et al. (2004) Stauffer & Grimson (1999) Elgammal et al. (2000) Comaniciu et al. (2003) Cavallaro et al. (2005)Maggio et al. (2007), where high-level events can be automatically detected, and multiple cameras can cooperate with each other, including different types of fixed cameras and mobile cameras hold by robots, as shown in Fig. 1. In the next-generation system in Fig. 1, namely cooperative visual surveillance network, new design challenges are introduced. First of all, the system configuration should be carefully designed since the distribution of the computations for these analysis functions will significantly affect the performances of these visual content analysis algorithms, and it will also affect the utilization efficiency of network resources. Moreover, the large computation of content analysis algorithms will increase the loading of servers, which will further limit the scale of camera number. Thanks to the advanced silicon manufacturing technology, which makes the transistor count in single chip increase dramatically, more and more functions can be considered to be integrated as a System-on-a-Chip (SoC) to cover more and more tasks for 7