This paper presents a video-based sleep monitoring system for real-life bedroom environments. We envision a smart living space with ambient intelligence components to record and monitor sleep in a non-invasive manner. We will describe the overall structure of the proposed system. Special emphasis, however, will be put on the formulation of a robust background modeling technique as the brightness in a real bedroom environment will change gradually at dawn, which creates a severe problem in video-based systems. To address the issue, we have demonstrated how to model brightness changes during morning hours. We have also investigated various texture-based background modeling techniques, including local binary/ternary patterns, to handle lighting changes at dawn in smooth regions for robust long-term monitoring. Comparative analyses have been carried out to examine the efficacy of different background modeling approaches on the sleep monitoring system.
Journal of Ambient Intelligence and Humanized Computing,4(1),57-66