A Distributed Multimodality Sensor System for Home-Used Sleep Condition Inference and Monitoring

In this paper, we propose a distributed system consists of heart-rate, passive infrared, and audio sensors for sleep condition inference. We apply machine learning methods to infer the sleep-awake condition during the time a user spends on the bed. This sleep-awake information would be useful for estimating critical factors including sleep latency, sleep duration, and habitual sleep efficiency related to sleep quality measurement. Our experimental results show that the proposed approach could be a good alternative to the traditional motion sensor Actigraph, with competitive performance on the sleep-related activity monitoring. Furthermore, the distributed computation nature of our system also makes it favorable for practical health-care applications