Activity-Aware Computing: Modeling of Human Activity and Behavior
Downey, Peter J.
Committee ChairDavies, Nigel
Downey, Peter J.
MetadataShow full item record
PublisherThe University of Arizona.
RightsCopyright © is held by the author. Digital access to this material is made possible by the University Libraries, University of Arizona. Further transmission, reproduction or presentation (such as public display or performance) of protected items is prohibited except with permission of the author.
AbstractWith our society becoming increasingly mobile and devices that are small, inexpensive and wireless, we are transitioning from an age of desktop computing to an age where computers are used in all aspects of life and leisure. Ubiquitous Computing is largely concerned with the progression of computers from stationary desktop environments to environments where computers and sensors are integrated with objects and every aspect of our daily life, often in an invisible way.This dissertation investigates an important problem in Ubiquitous Computing: detecting domestic activities using ubiquitously deployed sensors from data sets of limited size. The dissertation assumes that home environments in the next 20 years will support a wide range of sensing technologies that are built in smart appliances and the surrounding environment (e.g. RFID tags and readers, accelerometers, temperature sensors etc.). The dissertation also assumes that there will be an abundance of embedded CPU power in the environment that will enable fast and efficient spectral analysis and feature extraction from sensor signals. Using efficient wireless technologies such as the new Bluetooth Wibree protocol, these devices will be able to communicate their sensed data in an efficient way.Two approaches are presented for domestic activity recognition from wireless sensors. The first approach is rule-based and logical in nature and is suitable when sensor data is not present for training. Importantly, fuzzy distributions model the uncertainty and variability in expert knowledge. The second approach is probabilistic in nature and learns by observation without human intervention. This approach uses Bayesian Learning and is optimized to deal with sparse data sets (with hundreds of sensor readings and few instances of activities). Further, a case study is presented in which activity recognition optimizes energy consumption for wireless PC cards that results in significant energy savings.This dissertation concludes by highlighting major and minor results. A summary of the author's future and current research efforts is presented including the application of activity recognition in medical interventions and resource allocation problems.
Degree ProgramComputer Science