CONCLUSIONS
With the advent of 3D accelerometer MEMS chips that
are both small and power-efficient enough to be included in
wearable devices, long-term monitoring of sleep and wake
phases has become an attractive and cost-effective instrument
to complement traditional sleep lab studies using polysomnography. The systematic evaluation of algorithms that detect sleep
and wake phases in such accelerometer data is still lacking,
however, as current personal sleep devices and systems on
the market are closed-source and not meant to be clinically
deployed.
This paper contributes to such systematic evaluation
of detection algorithms by presenting a challenging and
publicly-available dataset5 with over 409 hours worth of
polysomnography-annotated 3D acceleration data at 100Hz for