We have presented a moving direction detecting system
using PIR sensors and machine learning technologies. To collect
PIR sensor signals, we have built a data collection unit
consisting of two pairs of PIR sensors and op-amp circuits,
and mounted it on the ceiling of a test room. Based on the
data set collected from six experimental subjects walking in
eight directions, we have performed classification analysis of
detecting the direction of movement using various machine
learning algorithms, considering the number of PIR sensors,
the field of view of PIR sensors, and the reduced feature sets.
Our results show that it is feasible to recognize the direction
of movement with 98% accuracy using the raw data set
collected from two orthogonally-oriented PIR sensors. We also
found that with the reduced feature set, we could also achieve
89%–95% accuracy according to machine learning algorithms