A Flexible and Scalable Architecture for Video Surveillance as a Service Systems

Abstract

In recent years, moving traditional video surveillance systems into the Cloud- based Video Surveillance (CVS) system or Video Surveillance as a Service (VSaaS) is significant to support a large number of IP cameras via the Internet. Most concerned about applying the cloud computing technology and quality of service rather than the scalability and flexibility of the system. In this thesis, the two issues are considered in the design and implementation of a VSaaS system architecture. The architecture addresses a flexible and scalable component- based VSaaS that can be easily scaled from one server up to a complex cluster to support the varying requirements of users. The publish-subscribe message passing mechanism has been used for the cooperation between the controller and compute node worker, and it contributes to the system fault tolerance and scalability. In case of cloud computing resource management in this architecture, cloud services are accessed via Amazon AWS, especially EC2 and S3 Application Program Interfaces (APIs) for computing services and object storage respectively, as many cloud computing providers are supporting those APIs. Moreover, this thesis also presents possible component deployment plans suitable for any size or type of systems, which combine both physical and virtual machines. The API server applying the REST interface and a token based authentication has been designed to support multiple types of clients as well as to protect the controller from direct security attacks. Also, this thesis presents the concept of having the compute node worker separately designed and worked apart from the video processor. Therefore, not only the compute node worker can support video processors but also related stream processing of which interfaces implemented based on standard I/O. In case of flexibility and scalability, the scheduling process plays an importantrole for the computing resource usage efficiency of a VSaaS system. Few previous works de- scribed video processing workload analysis and few video processing factors were revealed. Having unknown resource usage information of video processing it usually is difficult to search for an appropriate available computing node to assign for the requested video processing task. This thesis discusses in details about video processing workload characteristics applying various parameters, such as the type of video processing task, frame rate, frame size, and computing node specification. The analytical results have been applied to design the scheduler process to suitably place video processing tasks at different compute node specifications. This thesis proposes the video processing workload exploration for observing the capacity of available compute nodes by varying the parameters of related video processing tasks. The exploration data is then stored in a database to be used by the scheduler as the information for estimating the video processing resource usage of a new video processing task. The method and algorithm for estimating the resource usage of a new video processing task have been suggested, employ- ing both data from the video processing task exploitation and CPU scaling factor. Furthermore, this thesis suggests the scheduler’s criteria to assist the VSaaS administrator in optimizing the system resource usage suitable for the system type needed.