EMPD: An Event-based Multimodal Physiological Dataset for Remote Pulse Wave Detection (Part 7 of 9)

Note: Due to file size limits, this dataset is divided into 9 independent parts. (Note on filename: The physical archive files are named EPMD_Part_xx, which corresponds to this EMPD dataset). Remote photoplethysmography (rPPG) based on traditional frame-based cameras often struggles with motion artifacts and limited temporal resolution. To address these limitations, we introduce EMPD (Event-based Multimodal Physiological Dataset), the first benchmark dataset specifically designed for non-contact physiological sensing via event cameras. The dataset comprises 193 multimodal recordings (68 seconds each) collected from 83 subjects. For each recording, the data acquisition system simultaneously captures: Detailed Hardware Specifications: Wrist Event Streams: Captured using a high-speed event camera (Prophesee EVK4) focusing on the radial artery region, with a spatial resolution of 1280 × 720 and a timestamp precision of 1 µs. A low-power continuous-wave laser pointer (660 nm) is used as an active illumination source to enhance the event camera's sensitivity to pulse-induced micro-motions on the skin. Facial RGB Videos: Recorded using an industrial-grade RGB camera (FLIR BFS-U3-31S4C-C) at an original resolution of 2048 × 1536 and 30 Hz (downsampled and compressed to 640 × 480 for the released dataset). These are designed for traditional remote photoplethysmography (rPPG) extraction methods (e.g., CHROM, ICA, POS). Ground-Truth PPG: High-fidelity contact-based photoplethysmography (PPG) waveforms and heart rate (HR) recorded synchronously at 60Hz using a clinically validated contact-type finger-clip pulse oximeter (CONTEC CMS50E). Metadata & Labels: The dataset includes a global metadata file (EPMD_metadata.csv, available in Part 1) that provides comprehensive demographic and physiological state labels for all recordings.