A Deep Learning-Based Contactless Driver State Monitoring Radar System for In-Vehicle Physiological Applications

The recent advancements in vehicle automation - and associated shift from driving to non-driving activities - has increased the importance of in-vehicle driver monitoring that is safe, trustworthy and useable. Physiological measurement of driver monitoring (rather than just eye tracking) is a nascent approach gaining attention in the space of in-vehicle technologies; however, existing contact sensor-based approaches raise concerns regarding system usage, complexity and privacy. This paper presents research which developed a novel, contactless heartrate monitoring system for drivers using Frequency Modulated Continuous Wave (FMCW) short-range radars, which was validated in vehicular environments. A combination of signal processing and neural network methodologies, incorporating Long Short-Term Memory (LSTM), was adopted to mitigate the effects of body motion and other motion artifacts that cause noisy radar data. The neural network was trained on ground truth data collected in parallel using a medical-grade BIOPAC ECG system. Similarly, the results were validated and compared against this ground truth. The experimental results evidence that FMCW radars are a promising methodology for in-vehicular cardio physiological applications, displaying an overall accuracy of 93% in detecting drivers’ heartrate (HR) and inter-beat-interval (IBI). Additionally, there was no significant difference observed in the RMSE results for driving and non-driving conditions, evidencing that the methodology performed efficiently in both the conditions. This paper demonstrates the benefits of FMCW radars for contactless physiological driver monitoring applications within automotive domains, and beyond.