Real-time processing of media streams: a case for event-based interaction

There are many challenges in devising solutions for online content processing of live networked multimedia sessions. These include the computational complexity of feature extraction and high-level concept recognition, the massive amount of data to be analyzed under real-time requirements and the intricate correspondence between low-level features and high-level concepts. Our approach to these challenges is a distributed architecture consisting of interacting components encapsulating feature extraction and concept classifier algorithms. The purpose of the framework is to simplify the development of applications for the domain of on-line multimedia content processing. In this paper we focus on the architecture of the framework and argue that it fits well to the publish/subscribe interaction paradigm, leading to an event-based interaction model. Furthermore, we analyze different aspects of the application domain in more depth, such as requirements for scalability, reconfiguration, migration, event notification selection, filtering, and ordering. The main contribution of this paper is, that we for each aspect show how a suitable event notification service may satisfy the corresponding requirements. We also describe parts of a framework prototype. In particular we report on how the event notification service used satisfies the identified requirements.