Performance Evaluation of Visual Object Detection and Tracking Algorithms Used in Remote Photoplethysmography

Changchen Zhao, Peiyi Mei, Shoushuai Xu, Yongqiang Li, Yuanjing Feng; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2019, pp. 0-0

While most existing remote photoplethysmography (rPPG) approaches employ off-the-shelf visual object detection and tracking algorithms, these algorithms may not be well suited for rPPG problem. The detection and tracking algorithms are designed to be robust to fast deformations, non-distinctive color, fast translations, etc. while rPPG cares about background intervention, region consistency, smoothness of the traces, etc. Hence, there is a gap between a good detection and tracking algorithm and the rPPG measurement accuracy. This paper aims at bridging this gap by evaluating the performance of popular detection and tracking algorithms widely used in rPPG methods. We establish a processing pipeline and choose four detection and tracking algorithms. Experiments are conducted on two publicly available datasets and one self-collected dataset. We find three key factors that affect the rPPG accuracy: 1) stability of the tracking trajectory, 2) content consistency, and 3) robustness to deformation and fast translation. This study highlights the need for developing novel detection and tracking algorithms dedicated to rPPG and gives some useful insights.