Multi-Feature Transformer-Based Learning for Remote and Accurate Vital Sign Estimation Using mmWave FMCW Radar

Non-contact measurement of vital signs is an important issue in modern healthcare, aimed at enhancing patient comfort and convenience. Achieving sufficient measurement distance and adaptability to patient movement is essential for effectiveness. Current research covers measurement ranges between 1 and 3.5 meters. This study introduces a novel multifeature transformer-based super-resolution model. By inputting and encoding low-resolution sequences of distance, phase, and phase variation rates, the model upsamples these sequences, generating additional sample points and transforming the lowresolution data into high-resolution sequences. This process improves the quality and range of FMCW radar measurements by increasing the sample points, enabling the model to extract vital sign information from low-resolution sequences and produce accurate results. From the generated high-resolution sequences, vital signs such as heart rate and respiration are then extracted. Experimental results show that, at a distance of 5 meters, the system achieves 87.3 accuracy for heart rate and 89.1 accuracy for respiration in a walking scenario, outperforming existing solutions by 8.6 % and 6.5 %, respectively.