Non-Contact Photoplethysmography for In-Vehicle Health Monitoring With Seamlessly Integrated Photodetector Array and Hardware-Accelerated AI.

While non-contact physiological monitoring offers clear advantages over contact-based methods in dynamic environments such as intelligent vehicles, its practical deployment remains constrained by low signal-to-noise ratio (SNR) and susceptibility to motion artifacts. Here, we present an integrated in-vehicle photoplethysmography (PPG) health monitoring system that overcomes these limitations through two synergistic innovations. First, a well-developed, seamlessly integrated six-sector MXene-on-Si photodetector array (SI-PDs) with a high responsivity (∼0.72 A/W), paired with hardware-level signal conditioning, enables high-fidelity PPG acquisition under extremely weak signals. Second, a hardware-accelerated multi-scale convolutional neural network (CNN) performs robust, real-time extraction of blood pressure (BP) and heart rate (HR). Experimental validation demonstrates reliable PPG capture from challenging body sites, with beat-to-beat BP prediction achieving R2 >0.9 for both systolic and diastolic pressures, an FPGA core inference latency of 0.18 ms with an energy consumption of 0.059 mJ per inference. After the initial buffering, a sliding-window strategy enables continuous prediction updates at a configurable refresh rate. Real-vehicle tests at 40 and 80 km/h confirm the system's feasibility under moderate-speed driving conditions. This work establishes a pathway toward high-accuracy cardiovascular monitoring within intelligent vehicles, facilitating proactive health management and enhanced driving safety.