Eyesweb is being created as part of an ongoing project into extracting emotive content from sensors, primarily video cameras. The system is oriented toward extracting content from images through the construction of data flow like networks. Each network consists of a series of processing steps to filter, refine and extract information. The system is very similar to eyes in that processing operates on images, converting information in the images through different data abstractions and knowledge bases. This information can then be mapped to make media interactive.
This paper contains some interesting notes on extracting sensor data from video images:
1. The idea of extracting multiple pieces of information from sensor data is a key concept. They call these "OA's" or Observer Agents. Each observer agent is essentially an algorithm for calculating some useful knowledge about what is happening in the space.
2. Use of back-propagation to extract postural body positions. I wonder if this works with any degree of accuracy, I would like to see some data on accuracy.
The idea of an Observer Agent has some resonance with keeping aspects of the system as independent but cooperating elements.