Unobtrusive monitoring of cardiac activity during sleep is a key enabler for home-based screening of sleep disorders. Piezoelectric strips on a mattress can capture ballistocardiographic signals without any body-worn device, but their reliability across sleeping positions and mattress stiffness remains insufficiently characterized.
We present a signal processing pipeline for detecting beat-to-beat cardiac intervals from 16 equidistant piezoelectric strips spanning head to feet, validated against simultaneous ECG. The dataset comprises recordings from 7 healthy participants, each lying in four standardized positions for 3 minutes per position on up to four mattresses of varying stiffness (H1–H4), totalling 93 recordings. Cardiac detection uses template matching with dominant shape extraction; signals from all 16 strips are then fused into a single ensemble estimate weighted by per-strip signal quality. Detected intervals are matched to ECG intervals within a 30 ms tolerance. Performance is assessed by (i) the proportion of matched intervals per recording, (ii) the residual duration difference of matched beats as a precision metric, and (iii) the effect of mattress and body position on matching performance, evaluated with a linear mixed-effects model.
Across 93 recordings, ensemble fusion achieves a median interval-matching rate of 98.4% (IQR 93.0%–99.5%). The residual duration difference between matched piezoelectric and ECG intervals has a standard deviation of 7.09 ms, well within clinical tolerance for downstream HRV analysis. Mattress stiffness has no detectable effect on matching performance (largest pairwise difference, H1 vs. H4, +4.76 pp). Body position has a modest effect, driven by supine outperforming left-lateral by +7.40 pp; the other position contrasts are not distinguishable.
These results demonstrate that piezoelectric mattress sensor strips, combined with template-matching detection and quality-weighted ensemble fusion, deliver cardiac interval estimates that are simultaneously accurate, precise, and robust. This supports their deployment as a practical tool for long-term, unobtrusive cardiac and HRV monitoring.