Continuous Body Monitoring - Ambient Intelligence for Scientific Discovery - page 210

Fig. 6. VO2 metabolic cart in a lab setting.
The energy expenditure algorithm in the SenseWear system is multi-sensory.
There are limitations to single-sensor products (e.g. pedometers, accelerometers)
in deriving calorie expenditure. Such products are capable of seeing only one or
two-dimensional activities and can not differentiate between the myriad activities
performed throughout the day. The SenseWear system is a tested and proved
first-of-its-kind multi-sensory design that can intelligently recognize contexts and
apply the appropriate sensors to measure the activity being performed. This
innovation not only improves accuracy, but in applications such as HealthWear
allows the system to communicate the contexts (activities) that are most effective
for achieving the wearer's weight loss goals.
BodyMedia has already developed software that integrates the signals from
the sensors in the SenseWear armband and processes them using algorithms
that provide estimates of energy expenditure during different types of physical
activity. Energy Expenditure, often calculated in units of metabolic equivalents
(METS) for exercise physiologists, is a size and weight independent measure of
calories burned. Energy expenditure is predicted by first classifying the data into
an activity class (e.g., ambulatory exercise, biking-like exercise, other exercise,
or resting). An appropriate regression equation is then applied. The classifiers
and equations were tuned on indirect calorimetry data collected from subjects
performing the relevant activities (see Fig. 6).
Pedometer. An onboard algorithm in the armband calculates steps based upon
the high rate two-axis acceleration data stream.
Wearer's Context. Many SenseWear systems can predict when the wearer is
in a number of standard states, including lying down, sleeping, being restful,
traveling in a moving vehicle, walking, running, biking, and other types of exer-
cising. These categorizations can be important for other algorithms (e.g. energy