Remote assessment of the Heart Rate Variability to detect mental stress

Abstract

In the present paper, we introduce a new framework for detecting workload changes using video frames obtained from a low-cost webcam. The measurements are performed on human faces and the proposed algorithms were developed to be motion-tolerant. An interactive Stroop color word test is employed to induce stress on a set of twelve participants. We record the skin conductance and compare these responses to the stress curve assessed by a webcam-derived heart rate variability analysis. The results offer further support for the applicability of stress detection by remote and low-cost means, providing an alternative to conventional contact techniques.

Remote assessment of physiological parameters by non-contact technologies to quantify and detect mental stress states

“…The results presented in this study [33] demonstrate the feasibility of using the cardiac response derived from a lowcost webcam to assess mental workload changes. The processing methods are motion-tolerant and robust to light deficiency [30].…”

“…We have proposed [33] a new filtering technique that was developed to remotely and robustly recover the instantaneous heart rate signal concurrently to photoplethysmographic amplitudes fluctuations from video frames acquired by a lowcost webcam. Thus, a continuous wavelet transform filtering method was developed to precisely recover cardiac parameters of all participants.…”

Heart Rate and Heart Rate Variability Estimation in the Presence of Motion Artifacts

“…From studies of still subjects [34] the top candidates are the forehead and the cheek regions. The most frequently applied approach to finding the ROI is to use OpenCV face detection [35], [36], [37], [38], [39], [40], [41], [42] which generates a face box. To extract the more relevant portions of the face box, one can simply choose to use some portion of the width and height [35], [41], or apply a skin detection algorithm to find and apply a "skin mask" to the ROI [42].…”

“…The most frequently applied approach to finding the ROI is to use OpenCV face detection [35], [36], [37], [38], [39], [40], [41], [42] which generates a face box. To extract the more relevant portions of the face box, one can simply choose to use some portion of the width and height [35], [41], or apply a skin detection algorithm to find and apply a "skin mask" to the ROI [42]. The second most popular selection is the forehead rectangle [43], [44], in some cases subdivided into multiple regions [45].…”

“…The RGB camera output can be converted into luminance/chrominance component using color space conversion (CSC) algorithms, resulting in LUV or YCbCr formats [54]. In [42] for example, the U component (red to green) was used and it was found to be less motion sensitive than the green channel only.…”