to-D convertor); software components involve interpolation,
normalization, filtration, feature extraction, and
detection and classification.
2) We have built a prototype and used it to instrument an
experimental bed. We have used the experimental bed to
collect load cell signals from 30 subjects who make 27
different body movements during each experiment. We
have compared the detected body movements against the
ground truth observed captured by a video camera, and
found that the average error rate is 6.3%.
3) We have also used the same data to classify these
27 body movements into big movements (those that
involve the entire body) and small movements (those
that only involve one part of the body). We compare the
classification results against the ground truth observed
by a video camera, and found that the average error rate
is 4.2%.
The remainder of the paper is organized as follows. In
Section II, we describe the hardware system design of MotionScale,
and in Section IV, we describe MotionScale’s signal
processing algorithms. We present our evaluation setup and
experimental results in Section V. In Section VI, we summarize
the existing bed-mounted body movement monitoring
systems, and compare their pros and cons. Finally, we provide
concluding remarks in Section VII.