CUHeart: ECG-Grained Unobtrusive mmWave Cardiac Monitoring for Unconstrained Scenarios

Compared to short-term, contact-based clinical cardiac assessments, continuous and unobtrusive heart monitoring offers timely warnings and richer diagnostic information. To achieve this, heart monitoring systems must be contactless, unconstrained, and capable of fine-grained sensing. However, existing mmWave-based solutions often struggle with body motion interference and the trade-off between user freedom and signal fidelity, making it difficult to address all the three requirements simultaneously and enable seamless monitoring in everyday scenarios. To bridge this gap, we present CUHeart — an ECG-grained, contactless cardiac monitoring system based on mmWave radar, specifically designed for unconstrained daily scenarios. We constructed a complete system framework and built the first mmWave cardiac sensing dataset for indoor, unconstrained conditions. Based on this dataset, we conducted feasibility analyses to identify the challenges introduced by real-world body movement and proposed targeted signal processing solutions. In particular, a multi-factor signal quality evaluation method is introduced for beamforming optimization, and a deep network architecture of Filter Dictionary Transformer is designed for robust temporal feature extraction. Extensive evaluation, comparison, and ablation studies demonstrate that CUHeart achieves reliable performance under unconstrained conditions and for diagnostic support, with median errors of 24 ms for inter-beat interval, 2.5 bpm for heart rate, and promising results of ECG waveform reconstruction. The system enables unobtrusive heart monitoring in daily life, free from physical contact, strict constraints, or user cooperation, making it a truly practical solution for long-term cardiac health management.