Using Wearable Activity Type Detection to Improve Physical Activity Energy Expenditure Estimation

Abstract

Accurate, real-time measurement of energy expended during everyday activities would enable development of novel health monitoring and wellness technologies. A technique using three miniature wearable accelerometers is presented that improves upon state-of-the-art energy expenditure (EE) estimation. On a dataset acquired from 24 subjects performing gym and household activities, we demonstrate how knowledge of activity type, which can be automatically inferred from the accelerometer data, can improve EE estimates by more than 15% when compared to the best estimates from other methods.

Keywords: Algorithms; H5.m. Information interfaces and presentation (e.g., HCI): Miscellaneous; Human Factors; Measurement; Wearable; accelerometer; activity recognition; energy expenditure; health; physical activity; wireless.

Figure 1:

24 subjects participated in the experiment (circle=female, triangle=male)

Figure 2:

A person wearing the experimental setup.

Figure 3:

Example EE (measured and estimates) for 2 subjects performing routine 1 (A and B) and routine 2 (C and D). Figures 3A and 3C compare measured VO₂ to Freedson, Swartz, Hendelman and Crouter regressions. Figures 3B and 3D compare measured VO₂ to subject-specific perfect classification with fitted regression, perfect and automatic classification with custom MET lookup and perfect classification with MET lookup.