Multimodal energy expenditure calculation for pervasive health: A data fusion model using wearable sensors

Accurate estimation of energy expenditure during exercise is important for professional athletes and casual users alike, for designing training programs and meeting their fitness goals. However, producing an accurate estimate in a mobile, wearable health-monitoring system is challenging because most calculations require knowledge of the subject's movement speed. Though determining precise movement speed is trivial on a treadmill, inaccuracies of the sensors in a mobile system have a negative impact on the accuracy of the final energy expenditure estimate. In this paper, we propose a novel method to calculate energy expenditure using sensor fusion, in which data from multiple sensors is combined to formulate the result, based on a linear-regression model. We combine data from our wearable system with embedded pulse sensor and pedometer to produce an estimate that is far more accurate than possible with the pedometer alone, reducing our mean-absolute error by 64.3%. These results indicate that it is possible to obtain an accurate energy expenditure estimate in a multi-sensor system, even with affordable, low-cost, and pervasive components that may not be accurate individually.