Abstract
Contactless respiratory monitoring is increasingly relevant for remote healthcare, infection-control workflows, and continuous assessment in non-clinical environments. Thermal imaging provides an unobtrusive modality for respiratory sensing, yet most existing systems rely on high-frame-rate or radiometric cameras that limit scalability. This work presents a real-time, edge-deployable thermal monitoring system that performs respiratory-rate estimation, temperature extraction, and multi-person tracking using a low-cost FLIR C5 camera despite its low temporal resolution. The pipeline integrates YOLO-based person detection, persistent ID tracking, and a robust temperature-scale extraction method using OCR with temporal smoothing and outlier rejection. Respiratory signals are derived from nostril-level thermal fluctuations using motion-gated sampling, adaptive resampling, FIR band-pass filtering, and Welch spectral analysis, with a spectral-coherence check to suppress unstable estimates. The full system runs on an NVIDIA Jetson Orin NX with real-time visualisation and per-person logging. Preliminary results show that reliable respiratory-rate estimation is achievable under low-frame-rate constraints, demonstrating the feasibility of low-cost thermal sensors for scalable, privacy-preserving remote health monitoring.

This publication has 10 references indexed in Scilit: