ETD

Archivio digitale delle tesi discusse presso l'Università di Pisa

Tesi etd-04062020-115942


Tipo di tesi
Tesi di laurea magistrale
Autore
GIOIA, FEDERICA
URN
etd-04062020-115942
Titolo
Thermal Facial Analysis Under Acute Stress Stimuli
Dipartimento
INGEGNERIA DELL'INFORMAZIONE
Corso di studi
INGEGNERIA BIOMEDICA
Relatori
relatore Ing. Scilingo, Enzo Pasquale
tutor Dott.ssa Colantonio, Sara
Parole chiave
  • thermal imaging
  • mental stress
  • image registration
  • face tracking
Data inizio appello
24/04/2020
Consultabilità
Non consultabile
Data di rilascio
24/04/2090
Riassunto
This study aims to investigate the facial thermal response to mental stress in healthy subjects to gain a better understanding of the physiological reaction and to explore the use of thermal imaging in the field of computational physiology.
In modern society, the overall well-being seems to be constantly disrupted by stressful events, unpleasant or demanding situations and many studies have been proving that a long term stress response, known as chronic stress, can be the cause of serious diseases including mental illness, depression, anxiety and personality disorders but also cardiovascular health problems, immune system dysfunctionalities and gastrointestinal disorders. Thus, to overcome or prevent these extreme pathological effects caused by stress, a method to monitor it constantly is needed in order to give us a wake up call when necessary. A stress detection through biosignals is already possible, there are many physical and physiological changes that are measurable, but the way to acquire them is either invasive or uncomfortable to be done during the normal daily routine.
For this reason, the introduction of the use of a thermal camera to monitor the mood or the psychological state could be huge to support health management, since it is portable, contactless and unobtrusive.
For this study thirty-five healthy subjects were recruited, nineteen males and sixteen females, aged between twenty and fifty years old. Each subject completed an acute stress induction protocol, implemented as an Android Application. The protocol consisted in five experimental phases, alternating periods of rest (5 minutes) and periods of stressor tasks (3 minutes). The stressor tasks were the widely known Stroop Test and a Mental Arithmetic Task, both specifically modified in order to enhance their effectiveness in inducing a stress response.
The participants were characterized by their level of stress perception, assessed by the PSS (Perceived Stress Scale), and their level of clinical anxiety , scored by the BAI (Beck Anxiety Inventory). At the same time, their physiological body response was monitored by a self-reported stress assessment and two gold standard biosignals in stress recognition, which are the Galvanic Skin Response (GSR) and the Electrocardiogram (ECG).
The facial thermal patterns were extracted from an RGB-IR technology. In particular, the visible spectrum HD images were used to detect the regions of interest of the face and to track them throughout the acquired frames, after the registration got the RGB images projected in the IR coordinate system. The thermal signals, extracted from each selected ROI in the IR frames, were processed and analysed to find a set of features as reliable indicators of stress.
The thermal signal processing included an innovative outliers detector and artefact correction method, while the GSR and ECG underwent the standard processing algorithms prior the extraction of the features used, in the literature, as the ground truth in stress assessment.
A signed-rank Wilcoxon test, followed by the FDR correction, was applied to the features obtained from all of the acquired signals, to eventually highlight the significant difference between the stress tasks and their preceding rest periods. This resulted in a success of the induction protocol, as an increased sympathetic activity was assessed during the mental stressors and it also pointed out the associated decrease in temperature of many facial ROIs.
Moreover, the analysis of the correlation between the tonic component of the GSR signals and the thermal ones confirmed the possibility that the thermal signal, like the skin conductance (SC), could be a reliable stress detector.
Finally, one of the IR features was ranked third, preceded only by two GSR features, by an SVM classifier, implemented for explorative purposes, reaching an accuracy higher than 90%.
These results are promising and, increasing the cohort study, can lead to the development of an automatic stress recognition software, only based on thermal images.
File