Partial Information Decomposition Inspired Integration of Vision and Radar Signal for Robust Sleep Respiration Monitoring

Sleep monitoring, particularly for obstructive sleep apnea (OSA), conventionally relies on polysomnography (PSG) or wearable devices, which can be uncomfortable and impractical for long-term use. This paper presents a novel contactless sleep monitoring system that integrates radio-frequency (RF) radar sensing with near-infrared (NIR) vision technology. The core innovation of this system is the application of Partial Information Decomposition (PID) theory to quantitatively assess the unique, redundant, and synergistic contributions of both radar and vision modalities. Experiments conducted in a controlled sleep laboratory with multiple subjects demonstrate that the IR-UWB radar effectively captures respiratory micro-movements during periods of minimal body motion, while the NIR vision system provides complementary information during body movements. However, both modalities experience diminished effectiveness during excessive movement. Then based on the result, we design a PID-guided fusion method that enhances the robustness and accuracy of respiration monitoring, offering a promising step toward reliable, non-invasive sleep monitoring solutions.