VicarPPG-2 is a video and physiological signal (rPPG) dataset as introduced in:
Gudi A, Bittner M, van Gemert J. Real-Time Webcam Heart-Rate and Variability Estimation with Clean Ground Truth for Evaluation. Applied Sciences. 2020; 10(23):8630. [
MDPI] [
arXiv]
Specifically aimed at evaluating rPPG algorithms at estimating heart rate and short-term heart rate variability (which requires a minimum observation of 5 min), we recorded the VicarPPG-2 Dataset. In total, 10 subjects participated in the data collection. The male to female ratio was 7:3 with an average age of 29±5 years, and skin types ranging on a Fitzpatrick scale from II to IV.
Participants were asked to sit in front of a computer screen (∼1 m distance) on which the instructions were shown, a webcam was mounted on top of the screen and an LED ring lamp was mounted behind the camera. Screen brightness was reduced as far as possible to minimize the influence of screen light on the face. All videos were recorded using a Logitech Brio webcam at a fixed framerate of 60 fps using an H.264 compliant encoder (Microsoft Media Foundation), and stored in mp4 containers. The recording location was illuminated by natural ambient light in addition to a LED ring lamp (Falcon Eyes DVR-300DVC) to prevent strong shadows and influences of large changes in natural light. The ground truth signals were recoded in the form of synchronized ECG signals at 250 Hz sampling frequency obtained via an Arduino based ECG board (AD8232), and synchronized PPG signals at 60 Hz obtained via a pulse oximeter device (CMS50E) attached to the left index finger of the participant. The following four scenarios/conditions were recorded for each participant:
- Baseline: Participants sitting naturally while watching a relaxing video or reading an article on screen.
- Movement: Participants performing four different types of pre-planned angular body/head movements: turning head side-to-side (shaking), moving head up and down (nodding), a combination of head shaking and nodding (round), moving eyes while keeping head still, and naturally bobbing their heads while listening to music (dance).
- Stress: Participants playing a stress-inducing Stroop effect based game.
- Post-workout: Participants sitting unrestrained after performing fatigue-inducing physical workouts to induce higher heart rates.
Each condition was recorded for a duration of 5 min to allow for the computation of short-term heart rate variability, a total of 200 min of video were collected. Out of 400 collected ground truth files, two had to be removed from the dataset due to excessive finger movement in the PPG device and gradual detachment of the ECG ground electrode, leading to unusable signals. This dataset stands out as it was explicitly collected for RPPG purposes featuring 5-min long 60 fps camera recordings under various physical/physiological conditions, with simultaneous ECG and PPG ground truth recordings. Informed consent was obtained from all participating subjects.
