Project Description
The Cassandra project focuses on the automatic generation of audio/video feature descriptors. These descriptors can then be used in a wide variety of applications, such as search and retrieval tools for large audio/video databases, automatic chaptering of movies, automatic insertion or removal of commercial blocks in video content etc. The backbone on which the various descriptor generating algorithms created by the researchers run is formed by the Cassandra framework which allows for distributed streaming processing of audio/video sequences.

For the Cassandra project the involvement of MiPlaza has been the development of the Cassandra Streaming Framework and a set of supporting tools: The Cassandra GUI for visualization and annotation of automatically generated feature descriptors and ground truth, a benchmarking framework for generating algorithm recall/precision graphs and the Cassandra Demonstrator, a 6-LCD/20 PC demo setup running the Cassandra framework showing real time feature descriptor extraction on a live video stream.

The framework and demo has been implemented using a combination of C++ and Java, the GUI and benchmarking framework in C++. The team uses an agile software development process, which results in 'potentially shippable code' in small iteration increments (usually 2-3 weeks).

Currently, MiPlaza's involvement in the project is to further mature the Cassandra Framework, increasing portability, flexibility and speed.

Pictures

The Cassandra GUI application, with some of the generated feature descriptors visualized in the feature bars at the bottom and the 'parallel shot' edit dialog open in the foreground.

The Benchmarking Framework: Displaying the recall/precision graphs for various settings for the shot boundary and parallel shot detector algorithms.

Another view of the Cassandra GUI, showing the annotated video content available in the database in the top window, and an analysis graph in the bottom window.