Conditional dependencies between the human activities and different contexts (such as location and time) in which they emerge, are well known and have been utilized in the modern Ambient Intelligence (AmI) applications. But the rigid topology of the inference models in most of the existing systems adversely affects their flexibility and ability to handle inherent sensor ambiguities. Hence, we propose a framework for activity recognition suitable for a distributed and evolving smart environment. On the one hand, the framework exhibits flexibility to dynamically add and remove contexts through autonomic learning of individual contexts capitalizing the spatially distributed AmI infrastructures. On the other hand, it shows resilience to missing data by boot-strapping and fusing multiple heterogeneous context information.
[1]
Gary M. Weiss,et al.
Activity recognition using cell phone accelerometers
,
2011,
SKDD.
[2]
Qiang Yang,et al.
Activity recognition via user-trace segmentation
,
2008,
TOSN.
[3]
Dieter Fox,et al.
Location-Based Activity Recognition
,
2005,
KI.
[4]
Angelo M. Sabatini,et al.
Machine Learning Methods for Classifying Human Physical Activity from On-Body Accelerometers
,
2010,
Sensors.
[5]
Fariba Sadri,et al.
Ambient intelligence: A survey
,
2011,
CSUR.
[6]
Mario Cannataro,et al.
Protein-to-protein interactions: Technologies, databases, and algorithms
,
2010,
CSUR.