Sammendrag
The ever increasing access to computers with multimedia and good communication facilities calls for better and more complex services. Distance education, video conferences and distributed medical treatment shows that cooperation between people is also carried out on the Internet. Many people utilize the potentials of multimedia computers for developing new distributed solutions in which media analysis is applied. A distributed surveillance system can for example find the number of individuals in a building – a system much
used in banks and other buildings in need of strong security measures. This displays a need of a general framework for content analysis of media.
Distributed Media Journaling (DMJ) is a project at the University of Oslo that has defined an architecture to support journaling of media in distributed systems. Media journaling is an extended concept for involving real time content analysis and media annotation. The architecture will support communication, distribution, migration, modularity, flexibility and reconfiguration in journaling of media.
In this report, a prototype which implements some of the components in the DMJ architecture is developed to see if the above-mentioned requirements can be supported. It is especially the demand for a component model, the communication between the components and the migration that will be elaborated on in this report. The prototype is developed with the technologies Voyager Universal ORB and Message Bus (MBus). Voyager supports the component model and mobility with mobile agents. MBus supports exchange of messages between the loosely connected components with IP multicast, and does meet the communication requirements in an eventbroker. Video is sent as Real-time Transfer Protocol (RTP) data and is analyzed by the use of Java Media Framework (JMF) in the analysis components.
The results show that the migration of the analysis components in the network is faster than starting a new component. Nevertheless, the difference is small as JMF doesn’t support serializing of the processing object. It also turns out that it takes long time to load the classes in the Java Virtual Machine (JVM). The migration in itself does not take much time as the size of the file is relatively small. To sum up, the implementation of the prototype shows that all requirements for the DMJ framework can be supported by the chosen component technology.