Person-in-Bed Detection from Mattress-Integrated Accelerometers

Non-contact vital sign monitoring in healthcare settings often requires the detection of a person’s presence in bed to ensure spurious readings are not recorded. Towards this end, we develop a technique to detect the presence of a person in bed using mattress-integrated accelerometer readings. Since the same readings are also used for vital sign monitoring, the proposed algorithm can be easily integrated into existing systems without additional sensors. Our technique is based on a transformer model that processes the time-series accelerometer readings to give a binary output indicating the presence or absence of a person in the bed at each time stamp. We achieve > 98% accuracy and < 3s latency using a model with only around 3.6k parameters, which we believe is suitable for implementation on DSP hardware. Our code is available at https://github.com/bakhil/pib.