2) Data Normalization. We normalize the data by using
subject’s weight.Weight is also computed by our system.
Because the system aims to detect motions, it just focuses
on the segment of data that contains large changes
or oscillations. To achieve this, the system performs
Local Mean Removal to remove the constant value in the
load cell data by using a sliding window. we determine
the segment of data only contains large changes and
oscillations.
3) Data Filtration. We filter the data by low pass filter
with 10 Hz as a cutoff frequency. We remove the high
frequency spikes or noise by this filtration.
4) Feature Extraction. We investigate three different features
in this study, i.e., peaks in log-scaled sum of the
square of the data (Log-Peak), peaks in the energy of
the sum of the data (Energy-Peak), and valleys in zero
crossing of the sum of the data (ZeroX-Valley).
5) Motion Detection and Classification. Using these features,
we detect body movements and classify these
movements as big or small movements using a simple
thread-based scheme.