In sharp contrast to diet, where objective tools are rare, many objective tools are available for PA evaluation,
such as indirect calorimetry, doubly labeled water (DLW), pedometers, heart rate meter, and accelerometer.
However, diet and physical activity (PA) are both key factors in the etiology of obesity and other chronic diseases, a practical and objective tool to perform joint diet and PA evaluation in free-living individuals does not currently exist.
Often, diet and PA are studied separately without considering their correlations. It has been pointed out that this approach segregates and fragments the problem domain, causing concerns about the usefulness of study results.
In our approach, we first visually identify the real-life events (including sedentary events) based on the acquired multi-sensor data, and record their durations from the automatically saved time stamps.
Next, we search the PA compendium (a database available here)
to find the best match and then record the corresponding metabolic equivalent (MET) values. Finally, the caloric expenditure was calculated using an empirical formula based on the MET value and the resting metabolic rate (RMR) which require the gender, age, weight and height of the subject.
Mechanisms of measurement using eButton
Procedure for calculating calorie expenditure of PA. The circle with a cross denotes multiplication
Assuming that the PA compendium and the demographic information are both accurate, PA identification accuracy fully determines calorie expenditure accuracy.
Resting metabolic rate (RMR) based on Mifflin Equation
For Men: RMR = 10* weight (kg) + 6.25 * height (cm) 5 * age (year) + 5
Form Women: RMR = 10* weight (kg) + 6.25 * height (cm) 5* age (year) 161
Physical activity compendium - MET values
Metabolic Equivalent (MET)Values for Activities in American Time Use Survey (ATUS), see
detail
Automatic physical activity classification
Testing result in six categories of activities: sitting-up(SU),sitting-still(SS),walking(WK), bowing (BW),crouching(CR) and waist exercise(WE).
Horizontal axis: resolution of motion orientation.
Vertical axis: recognition rate using naive Bayes, K nearest neighbors and support vector machine.
Image features: our multiresolution good feature detection method (MRGF).
Multitouch technology for PA categorization