Stress Estimation Using Multimodal Biosignal Information From RGB Facial Video

Takumi Nagasawa, Ryo Takahashi, Chawan Koopipat, Norimichi Tsumura; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2020, pp. 292-293

In the present paper, we propose a method for acquiring multiple biological information inputs from a red-green-blue (RGB) facial video footage and using their feature values to estimate stress levels. Such estimations are important because if left unchecked, stress can cause severe mental illness and/or physical damage to the human body. Accordingly, it is important to understand the onset of stress at an early stage and take measures to counteract it. However, since it is difficult for us to accurately gauge our stress levels, it would be desirable to establish an objective and accurate estimation method. Additionally, while the commonly used questionnaire method is easy to implement, it lacks both objectivity and accuracy. In a recent study, many methods that use biological information were proposed. In the present study, we estimate stress using three biological signals captured using an RGB camera: pulse, blinking rate, and pupil diameter. Our results show that stress estimation accuracy is improved by using these biological signals, thereby indicating that it is possible to estimate stress more accurately by using biological information in a multimodal manner.