Computer Science Colloquia
Tuesday, January 22, 2013
Advisor: John A. Stankovic
Attending Faculty: Kamin Whitehouse (Chair), Alfred Weaver, Stephen Patek, and John Lach (minor representative)
2:00 PM Rice Hall, Rm. 242
PhD Proposal Presentation
Realism in Activity Recognition for Long Term Sensor Networks Deployments
Research in wireless sensor networks has been very successful in creating testbeds and short term (typically less than one or two months) real deployments for many application areas (e.g, home health care, saving energy in buildings, infrastructure monitoring, agriculture, and various environmental science applications) that depend on accurate activity recognition. However, for many real home situations these activity recognition solutions are not robust enough for long term deployments. The reason is the presence of a form of phase transition in long term deployments that results into many realities that are not always seen in short term deployments. Another realism of these deployments is the variability in different smart home environments. Existing systems are designed based on some assumptions about the environment and need significant modification to be applied in other environments. We propose to develop robust and reliable activity recognition for in-home deployments that address these realities of long term deployments including difficulty in obtaining labeled ground truth for training activity recognition systems, individuals performing overlapping activities, generalizability in varying smart home environments, and handling the evolution of human behaviors. In addition, the utility of activity recognition often depends on recognizing anomalies from typical behaviors. Anomaly detection can also suffer from the realisms of long term deployments. Therefore, we also propose a novel semantic anomaly detection system in human behavior.