Periodic clinical visits are merely “snapshots” of the patient's status, giving rise to portable monitoring instruments. This thesis describes a second generation system for objective and unobtrusive physical activity monitoring. The approach recognizes that complex dynamic situations can be characterized by activity states defined by feature combinations.; A generic platform supporting research and field use was developed for a subject-worn system incorporating a MEMS-based sensor, new microcontroller technology, and a “raw data” logging mode for data-basing sensed parameters, An experiment was conducted with 10 subjects and five test states: lying, sitting, standing, walking, and running. Data representing thigh acceleration was logged and processed to derive amount of motion and orientation. Histograms were used to develop state detection rules found to be approximately 95 percent accurate across all states, It is concluded that the system and approach will support exploration of a wide variety of activity states and detection schemes. |