Multi-Body, Multi-Function Body Sensor Networks

Li, Qiang, Computer Science - School of Engineering and Applied Science, University of Virginia
Stankovic, John, Department of Computer Science, University of Virginia

With the cost of health care increasing, Body Sensor Networks (BSNs) have been proposed to provide low-cost health care solutions for the senior population. Reliability is essential for BSNs because they are mainly used for medical purposes such as detecting health-detrimental accidents and monitoring health status. The reliability of BSNs is twofold: reliable communication and reliable applications. The former focuses on data collection, while the latter focuses on data processing. Though existing work studies some of the reliability issues, the state of art lacks systematic approaches to implement reliable BSNs, especially when multiple BSNs coexist.

Therefore, in this dissertation we systematically study how to develop reliable multi-body, multi- function BSNs. First, we perform an empirical study to investigate the challenges for developing reliable BSNs. Then, we propose a QoS framework that guarantees reliable communication and fidelity of BSN applications. Reliable communication is achieved by dynamically grouping nearby BSNs into one group and using different frequency channels and time sharing to schedule transmission within each group. Fidelity of BSNs is guaranteed by profiling BSN platforms and applications, and finding the optimal configuration to accommodate multiple BSN systems. We also develop two representative BSN applications, fall detection (accident detection) and social activity detection (health monitoring), to study how to develop reliable BSN applications. At last, we build a BSN platform to unify our proposed QoS framework with both fall detection and in-person interaction monitoring applications.

The evaluation demonstrates the effectiveness of our proposed methods for developing reliable multi-body, multi-function BSNs. BSN systems implemented using our framework achieves over 97% overall Packet Reception Ratio (PRR) even when multiple BSNs are within the interference range of each other, while PRR of systems not considering nearby BSNs drops to below 50%. Profiling BSNs and automatically switching between BSN settings guarantee data fidelity in terms of sample frequency and sample delivery. Our fall detection application detects over 96% of falls and our in-person interaction monitoring application achieves over 82% accuracy.

PHD (Doctor of Philosophy)
Body Sensor Networks, BSN, QoS, Fall Detection, Social Interaction Monitoring
All rights reserved (no additional license for public reuse)
Issued Date: