Competitive State Anxiety Inventory Assessment using Remote Photoplethysmography and Deep Learning

Abstract

For measuring human anxiety Competitive State Anxiety Inventory (CSAI) is a very popular method. Though this method is widely used in determining anxiety in the sports area, recently, this method is also deployed in other activities. Our objective was to determine the CSAI-based assessment non-invasively using HR-HRV for three different tasks. A baseline video and a target video of the same subject are used and the outputs are the competitive state anxiety inventor are proposed in this study. The framework uses a support vector regression, based on the abstract features derived from the deep learning, heart rate and heart rate variability features from the remote photoplethysmography signal extracted from the facial videos. Deep learning could provide good estimate results. The results proved the rPPG features could improve the accuracy of the cognitive anxiety and somatic anxiety prediction. The competitive State Anxiety Inventory (CSAI) can be assessed by HR-HRV analysis from the facial video. The machine learning algorithm can predict human anxiety in CSAI units despite sports activities. The result from our framework (StressFusion) shows a strong correlation between the ground truth (from UBFC-Phys dataset) and predicted anxiety scores.