Publications

The Heart Rate (HR) is a vital sign that is used to assess the physical and mental state of an individual. There is a growing interest in incorporating HR measurement into Driver Monitoring Systems (DMS), providing physiological measurements to help address long-existing road safety issues by minimising human error.

In real-world driving scenarios, the HR must be measured using non-contact approaches that avoid distracting or restricting the driver. The most common approaches to non-contact HR measurement use either computer vision (CV) or mm-wave radar, both showing acceptable performances in controlled studies. However, the relative merits of different sensor modalities for real-world scenarios remain unclear, and the potential benefits of a combined approach are unquantified. 

To address these questions, this paper first proposes and implements non-contact HR measurement architectures for both CV and mm-wave radar systems and characterises their HR estimation performance, using electrocardiography (ECG) to provide ground truth measurements. The effects of distance to sensors and of illumination variations on HR estimation are also studied, showing the relative errors for both modalities to be less than 0.5% for the distances found in practical DMS. These results also highlight the distinctive characteristics of each modality and the benefits of a multi-modality approach for DMS.

Advances in noncontact vital sign detection have demonstrated its vast potential to supplant conventional contact sensors, not only within the established healthcare system but also across emerging domains such as smart homes, security systems, and in-cabin sensing. Noncontact cardiorespiratory measurements are among the most common, in part due to their relative ease of measurement using noncontact sensors.

For this application, radar-based sensors have several advantages, including high accuracy measurement that is invariant to ambient lighting conditions, which are especially beneficial in non-clinical settings. Radar-based cardiorespiratory feature extraction architectures consist of two parts: the radar hardware design and the signal processing pipeline. However, the combined complexity of the hardware, the underlying physics and the scene itself, makes the signal processing requirements very challenging.

To address this issue, we first review the recent trends in this domain, including the move towards Frequency Modulated Continuous Wave (FMCW) radar sensors, and then present an empirical investigation using a commercial FMCW radar to illustrate the unsolved real-world signal processing challenges for noncontact cardiorespiratory measurement. Additional in-depth analysis is used to interpret the underlying reasons behind the challenges, together with potential solutions.

The work presented will benefit researchers and industrialists working on radar-based physiological measurement, facilitating a greater understanding of the problem, its benefits and challenges, and potential future research directions.

Road safety remains a critical challenge, with human errors such as drowsiness and distraction as contributing factors. Physiological indicators such as Heart Rate (HR) and Respiratory Rate (RR) provide insights into understanding the driver state but are difficult to acquire in real-world driving scenarios due to the constantly changing driver pose and complex ambient environment.

To tackle this challenge, a noncontact cardiorespiratory monitoring system using Frequency Modulated Continuous Wave (FMCW) radar is proposed. Unlike optical or acoustic systems, the use of radar can minimize interference from the ambient environment. However, its relatively low spatial resolution, driver motion, and variability in human Radar Cross Section (RCS) present significant signal processing challenges, especially for target tracking.

To address these issues, a novel target tracking algorithm and a reference-free Signal Quality Index (SQI) based on phase continuity are proposed, aimed at significantly improving the consistency and reliability of target tracking and overall signal quality. A performance evaluation using a driving simulator and synchronized Electrocardiography (ECG) recordings for ground truth demonstrates that our proposed method achieves a mean absolute error (MAE) of 1.65 beats per minute (BPM) for HR, 0.79 respirations per minute (RPM) for RR, and a relative accuracy of 97.83 % and 95.06 %, respectively.

These results highlight the potential of the proposed system to enhance driver monitoring by providing accurate and continuous HR and RR measurement, contributing to the early detection of adverse driver states.