The success of contemporary surveillance systems depends on their cost and their ability to automatically and correctly interpret events that take place in an observed environment. Correct decisions enhance the situation awareness of an operator, while wrong ones lead to false or no alarms, damaging the system's credibility. In this paper a low-cost, secure, cloud-based surveillance system is described for monitoring wide dynamic environments. The focus of our current work is to present the system's overall architecture and discuss in detail the design and functionalities of its main modules: (a) the multiocular network of multispectral cameras for visual information gathering, (b) the intelligent segmentation, event modeling and anomaly detection module and (c) the end-to-end power-aware data encryption infrastructure. By following the proposed approach, different geographic areas can be successfully covered, high-level situations of interest can be modeled and secure data communication can be guaranteed, at a minimum necessary electrical and CPU power. Finally, an application scenario in the maritime domain is provided.