Behavior modeling and recognition methods to facilitate transitions between application-specific personalized assistance systems

Abstract

Activity recognition mandates complex sensor fusion processing. Many contributions in the literature focus on improving the recognition accuracy of a limited set of activities or the efficiency of the algorithms. However, there is little work on how to dynamically adapt the activity recognition techniques when evolving from one situation to the next. We present tool support to model transitions between activities, and a modular distributed framework of human activity recognition components with support for analyzing resource and recognition tradeoffs for different deployments and configurations.