Pattern analysis for embedded intelligence

The space-time processing approach investigated by the group is one appropriate strategy for robust analysis of multiple sensor data encompassing dynamic processes such as motion, variable shape, and appearance, whereas traditional time-based approaches require additional modeling tools (e.g. Markov chains) for dynamical processes. In the development of methodologies for the space-time domain over the last two decades, the research focus has mostly remained on the development of low-level cues, which have incrementally become more descriptive (e. g. transition from simple motion cues to space-time shape).

Frameworks of pattern analysis algorithms proposed in recent years attempt to tackle increasingly complex scenarios. Complexity can be measured by the fact of how difficult is to obtain an automated high-level representation for a given scene using a network of heterogeneous sensors. The ultimate goal to make pattern analysis applications become part of our daily lives imposes the requirement that sensors must operate reliably nearly anytime and anywhere; a requirement which brings forth variability in the sensor data content. Sensing - both in humans and computers - is a dynamic process, thus variations always involve the spatial and temporal dimensions.

The research group aims to contribute to a deeper understanding of the unified processing of spatiotemporally generated sensor data and of its space-time structure in special as the basic knowledge needed to stimulate embedded/ambient intelligence.