| Stress is a major health concern that not only compromises our quality of life, but also affects our health and well-being. Despite its importance, our ability to objectively detect and quantify stress in a real-time, non-invasive manner is limited. This capability would have a wide variety of medical, military, and security applications. Under the dissertation research, we have developed a pipeline of image and signal processing algorithms for such a capability, which includes remote blood wave (BW) signal detection based on minor color intensity fluctuations in visible spectrum videos of the human skin during the cardiac cycle, and physiological stress measurements based on the temporal variability in these remotely detected cardiac signals.;To detect BW signals reliably, we applied Principal Component Analysis (PCA) for noise reduction and Independent Component Analysis (ICA) algorithms for source selection. A blind deconvolution (BDC) algorithm based on least squares (LS) minimization was then applied to the BW signals to determine peaks, which were then used to detect reliable RR-intervals, the intervals between the adjacent BW signal peaks. The series of RR-interval values was then used to derive heart rate variability (HRV) features in both temporal and frequency domain. The stress significant features were then identified based on the paired t-test, which were then used with logistic regression (LR) classifier to discriminate stress states from the normal ones. In addition, we have defined a new metric called differential pulse transit time (dPTT) as the difference in arrival time of BW signal at two separate distal locations, and have demonstrated its potential use for stress detection.;The developed algorithms were tested against the human subject data collected under two physiological conditions using the modified Trier Social Stress Test (TSST) and the Affective Stress Response Test (ASRT). This dissertation presents the developed algorithms and the stress detection results. This research provides a proof that the variability in remotely-acquired BW signals can be used for stress (high and mild) detection with 92% likelihood of being correct, and a guide for further development of a real-time remote stress detection system based on remote HRV and dPTT. |
