PhysDrive is the first large-scale, multimodal dataset for in-vehicle non-contact physiological measurements in a real-world driving environment, released in 2025 by institutions including the Hong Kong University of Science and Technology (Guangzhou), the Hong Kong University of Science and Technology, and Tsinghua University. The related research paper is titled "...".PhysDrive: A Multimodal Remote Physiological Measurement Dataset for In-vehicle Driver MonitoringThe data has been selected for NeurIPS 2025 Datasets and Benchmarks, aiming to support the research and evaluation of driver condition monitoring, smart cockpit systems, and multimodal physiological perception methods.

This dataset contains approximately 24 hours (1500K frames) of multimodal recordings from the real driving processes of 48 drivers, covering different lighting conditions, vehicle types, driving actions, and road conditions. It simultaneously acquires three types of sensory modalities: RGB video (30fps), near-infrared video (30fps), and millimeter-wave radar data (20fps). The dataset also provides six types of physiological ground truth signals, including electrocardiogram (ECG), blood volume pulse (BVP), respiration (Resp), heart rate (HR), respiratory rate (RR), and blood oxygen saturation (SpO₂).

Data distribution:

Each participant completed 6 driving segments, each lasting approximately 5 minutes, for a total of approximately 30 minutes and 16.5 km.

• Lighting conditions (4 categories)
  • Midday stable light
  • Unstable light in the morning/evening
  • Cloudy/Rainy day with low light
  • Low light at night
• Vehicle type (3 categories)
  • A0 class vehicles
  • Class B vehicles
  • C-segment SUV
• Driving maneuvers (Class 2)
  • stationary driving (no additional actions)
  • Interactive/speaking state (increased natural movements)
• Road conditions (Class 3)
  • Flat and barrier-free roads
  • Flat but congested roads
  • Bumpy and congested roads