## Overall ##
This dataset contains data to evaluate and train remote photoplethysmography (rPPG) methods to estimate heart rate remotely by processing video that contains facial area (see [1]).
The data are of two types:
- source data obtained in the experiments: color signals acquired from videos (*color_signals* folder) and heart rate values measured by means of pulse oximeter (*ground_truth* folder);
- RMSE values calculated for different pipelines of rPPG method described in [2] (*pipeline_rmse* folder).
Numbers in file names are the experiment identifiers. The same numbers correspond to the same experiment.
A detailed description of the process to obtain the data can be found in [1, 2].
## Color signals ##
The color signals are obtained from videos containing person's face by averaging red, green and blue color components within certain area with coordinates set relatively to rectangle of face detected in each frame. The face is detected by OpenCV implementation of Viola-Jones algorithm [3].
Color signals are grouped in folders Log, Mic, Len in accordance with cameras used for their capturing.
The structure of each file with color signals is as follows. The first 21 lines contains meta information of the file (to keep “matrix-style” structure, spaces are replaced by zeros in meta information section):
- the 1st line contains experiment id (identifier), total number of frames in the video and video duration (in seconds);
- the 2nd line contains number of areas used to extract the color signals from facial region, number of areas to extract the color signals from background region, and technical value that should be skipped;
- each line from 3rd to 20th contains 4 numbers (x, y, width, height) defining relative coordinates of areas within the facial rectangle to extract color signal;
- the 21th line contains technical value that should be skipped.
The remaining lines contain average values of red, green, and blue color components within the corresponding area. Each line corresponds to the consequent frame.
## Ground truth ##
Heart rate values for each experiment were obtained by means of the Choicemmed MD300C318 pulse oximeter for finger (with declared accuracy 1-2 beats per minute).
Heart rate calculation by the pulse oximeter was started synchronously with starting of video recording. The pulse oximeter wrote heart rate values every 4 seconds.
The structure of ground_truth files is as follows. The first 16 lines is a technical unimportant information. Next, in each line the last value represents the heart rate calculated as average heart rate within the 4 previous seconds. (Another value represents oxygen saturation.)
## Pipeline RMSE values ##
Each file corresponds to the certain camera (CamLog/CamMic/CamLen) and motion type (Stat/Dyn) [2]. In files, lines correspond to experiment ids, columns correspond to rPPG pipelines described in [2]. The RMSE values are calculated between reference heart rate values (ground truth) and pipeline estimations.
## References ##
1. M. Kopeliovich and M. Petrushan, “Color signal processing methods for webcam-based heart rate evaluation,” In: Bi Y., Bhatia R., Kapoor S. (eds) Intelligent Systems and Applications. IntelliSys 2019. Advances in Intelligent Systems and Computing, Vol. 1038, p. 703–723, 2020.
2. Kopeliovich M., Mironenko Y., Petrushan M., “Architectural Tricks for Deep Learning in Remote Photoplethysmography,” Proceedings of the IEEE International Conference on Computer Vision Workshops, 2019.
3. Kopeliovich M., Petrushan M. (2018) Approximation-based transformation of color signal for heart rate estimation with a webcam. In: Proceedings of the International Conference on Pattern Recognition and Artificial Intelligence, pp 638–642.
4. OpenCV (2013) OpenCV 2.4.6.0 documentation. URL http://
docs.opencv.org/2.4.6
