Contactless Smart Infant Sleep Monitoring System

Developed a multispectral physiological imaging system for precise, contactless monitoring of infant vital signs, in collaboration with a leading hospital in Wenzhou.

Established an interpretable video-based sleep/wake classification model to enhance monitoring accuracy from consumer-grade to clinical-grade standards.

Leveraged ECG and PPG signals to extract and validate infant motion metrics for preliminary sleep analysis.

Deployed open-source human pose estimation and optical flow algorithms to compute key-point motion features, confirming feasibility of camera-based polysomnography (PSG) for infant sleep staging.

Conducted camera-based PSG monitoring on 100 infants to establish normative sleep-stage benchmarks.

Extracted limb movement coordination and intensity features from video data, applying SVM and Random Forest classifiers for binary and multi-class sleep-stage classification.

Applied DL algorithms (LSTM, Transformer) to improve the accuracy of infant sleep stage classification.