Fig. 1: OpenMetabolics energy expenditure estimation method.
From: OpenMetabolics: Estimating energy expenditure using a smartphone worn in a pocket
a A participant walks with a smartphone in their pocket. b An orientation calibration algorithm aligns the smartphone data with the thigh’s frame of reference during each bout of activity, regardless of the smartphone’s orientation in the pocket. c Individual gait cycles are segmented by detecting peaks in the sagittal plane angular velocity (ωz). d Each axis of angular velocity data from a single gait cycle is downsampled to a fixed size of 30 values. e A linear regression model estimates motion artifacts caused by the phone shifting in the pocket during walking and removes these artifacts from the uncorrected gait data. f A pre-trained data-driven model, an ensemble of gradient boosted trees, estimates the energy expenditure once per gait cycle. This model takes an input of the corrected gait data, statistical features of the gait data, and the subject’s height and weight. Each tree estimate (p) is aggregated using a learning rate (λ) to produce the ensemble estimate. g OpenMetabolics estimates energy expended during real-world activities.
