Classifying Physical Activities from Accelerometry data recorded from healthy, elderly and neurological subjects

By 2050 two billion people will be aged 60 or older representing 22% of the global population. As this skew in the population distribution develops the nature of elderly health care provision will need to evolve. This dissertation investigates the practical application of accelerometer based sensors as inexpensive tools to monitor various physical activities and neuro-degenerative disorders. Four distinct studies were conducted. The first and second study investigated the ability of supervised learning algorithms to classify various physical activities. Specifically, several classifiers were trained on time-domain and frequency-domain features derived from accelerometry data gathered from healthy and elderly populations in a laboratory and home environment, respectively. The primary findings were that a correctly parameterized supervised classifier could achieve high sensitivity and specificity in activity recognition without the need for user specific training data while only employing sensors affixed to the wrist, arm and chest. In the third study conducted a body sensor network to detect motor patterns of epileptic seizures was developed. A template matching algorithm based on accelerometry data was successfully designed to monitor seizure occurrence outside the laboratory setting. Finally, the fourth study conducted investigated the capacity of an accelerometer based sensor affixed to the upper chest to detect gait and balance impairments in pre-symptomatic and symptomatic Huntington¿s disease subjects. The primary motivations were the known limitations of commonly used ordinal based clinical tests and the considerable expense of laboratory-based walkways and other quantitative systems currently employed to monitor Huntington¿s disease. By analyzing spatio-temporal gait parameters derived from the accelerometry data Hunington¿s disease progression was identified. These studies demonstrate the practical application and indicate the significant potential of accelerometer based sensors as inexpensive tools to monitor various physical activities and neuro-degenerative disorders