| Physical activities (PA) in the form of walking, jogging, sport activities, etc., is essential to maintaining health and preventing cardiovascular disease, diabetes, and obesity. Accurate assessment of PA is critical to quantitatively assessing the type, intensity, and duration of physical activities that a person has engaged in, and is therefore of great interest to the medical and biological research community. Accurate PA assessment requires addressing three challenges: (1) how to distinguish different types of activities that produce similar motion profiles but have different energy expenditure; (2) how to interpret and integrate different sensor data for accurate identification of both the type and intensity (i.e. energy expenditure) of PA; (3) and what features to extract from the sensor measurements to enable computationally efficient PA classification. The presented work addresses the above challenges through research on the following three issues: 1) Concurrent design, modeling, analysis, and prototyping of a multi-sensor, wearable integrated measurement system (IMS) with wireless data transmission capability. 2) Nonparametric, discriminative data classification and regression for multiple sensing data fusion. 3) A featureless classification framework through sparse representation for PA type classification.;The wearable sensing electronics and computational method developed in this work contribute to advancing the state-of-the-art in human PA assessment by 1) enabling real-time physical activity measurement in free-living conditions, 2) improving the fundamental understanding of machine learning for PA modeling, and 3) providing more accurate PA assessment than traditional methods. |
