Addressing Reproducability Issues in Remote Photoplethysmography (rPPG) Research: An Investigation of Current Challenges and Release of a Public Algorithm Benchmarking Dataset.
Date created: | Last Updated:
: DOI | ARK
Creating DOI. Please wait...
Description: In current study the needs in rPPG research and development were investigated and it was concluded there is a serious lack of reproducibility and comparability in the current rPPG research. A public benchmark dataset was created and the first rPPG algorithm, a testing tool by VicarVision, was benchmarked. Three current main challenges for rPPG were defined and incorporated in the dataset; lighting in combination with skin tone, motion robustness and high heart rates & pulse-rate change robustness. Videos accompanied by ECG measurements were recorded on three participants with three different skin tones. In the test of VicarVision’s rPPG tool it was found that the tool performs very well for bright skin tones in different lighting conditions as well as for high heart rates. Motion and skin tone robustness however are to be improved. Repetitive motions can trick the algorithm to be perceived as the heart rate, and the algorithm performs significantly worse for dark skin tones. Possible explanations can be found in the physical properties of darker skin or in a possible lack of dark skinned subjects in the training data. The large performance gap between dark and bright skin tones forms a challenge for future rPPG algorithms and it is crucial this challenge is properly addressed before real life applications can properly be introduced.