MSPM: A Multi-Site Physiological Monitoring Dataset for Remote Pulse, Respiration, and Blood Pressure Estimation

TL;DR

MSPM addresses the need for ground-truth, multi-site camera-based physiological sensing by providing 103 participants with full-body RGB video, multi-site PPG ground truth, cuff BP, SpO2, and respiration, plus multi-angle NIR data. The dataset enables canonical studies in multi-site rPPG, PTT, and remote respiration, supported by robust signal fusion and ROI strategies. Key findings show intra-dataset pulse-rate MAEs below $4$ BPM and cross-dataset MAEs reaching as low as $2$ BPM in some cases, respiration MAE around $1.09$ breaths per minute, and strong correspondence between remote and contact PTT signals, suggesting practical utility for camera-based BP research. MSPM’s breadth and adversarial-attack evaluation offer a valuable resource for advancing non-contact vital-sign estimation and cross-domain generalization in real-world scenarios.

Abstract

Visible-light cameras can capture subtle physiological biomarkers without physical contact with the subject. We present the Multi-Site Physiological Monitoring (MSPM) dataset, which is the first dataset collected to support the study of simultaneous camera-based vital signs estimation from multiple locations on the body. MSPM enables research on remote photoplethysmography (rPPG), respiration rate, and pulse transit time (PTT); it contains ground-truth measurements of pulse oximetry (at multiple body locations) and blood pressure using contacting sensors. We provide thorough experiments demonstrating the suitability of MSPM to support research on rPPG, respiration rate, and PTT. Cross-dataset rPPG experiments reveal that MSPM is a challenging yet high quality dataset, with intra-dataset pulse rate mean absolute error (MAE) below 4 beats per minute (BPM), and cross-dataset pulse rate MAE below 2 BPM in certain cases. Respiration experiments find a MAE of 1.09 breaths per minute by extracting motion features from the chest. PTT experiments find that across the pairs of different body sites, there is high correlation between remote PTT and contact-measured PTT, which facilitates the possibility for future camera-based PTT research.

Figures (16)

• Figure 1: Key novelty of MSPM: Ground truth PPG signal collected across multiple sites on the body with full-body video for its full utilization.
• Figure 2: Placement of sensors in the collection room.
• Figure 3: PPG sensors were attached to the subject's neck and limbs.
• Figure 4: Imaging modalities in MSPM include RGB video from the front \ref{['fig:rgbfront']}, overhead \ref{['fig:rgbtop']}, and profile \ref{['fig:iphone']}, and NIR video of the eyes \ref{['fig:nir']}.
• Figure 5: Duration of each activity, averaged across sessions. The Hand Raise, Relax segments, and videos (Respiration, BW Video, Color Video, Adversarial Attack) automatically advanced on completion thereby resulting in low duration variability.