A Software Architecture for Video Analytics

Video analytics is the automatic understanding of complex events occurring in a captured scene, by artificial intelligence. It is a fundamental task for homeland security and crime prevention. Some video analytics tasks are being addressed by at the edge processing approach, which combines algorithmic with hardware development, within the camera system design. The at the edge processing approach looks for real-time performance on specific domains and low network bandwidth requirements. However, it implies that the already installed city-camera-network should be replaced and continuously updated since it introduces constraints to analytics capabilities related to cameras’ characteristics and functionalities. Thus, city surveillance by the at the edge processing approach requires a large budget, and continuous investment, to keep the system working. In this paper, we present both faced challenges and obtained achievements during the design process of video-analytics-system software architecture. Interoperability, availability, and security were the prioritized quality attributes by the Quality Attribute Workshop method. A microservices and cloud-based design was the result of applying the Attribute Driven Design method, incorporating software engineering sustainability concerns. As a practical advantage, the designed system can work with the already installed city surveillance camera network and allows incorporating video analytics algorithms as the surveillance system evolves. The evaluation combined the functional and operative prototype as well as the ATAM method, according to which approach is more suited to evaluate a specific driver. The discussion on findings of design an evaluation processes is a takeaway for the reader.