• affective computing
• approximate entropy
• Biomedical signal processing
• Bispectrum
• Cardio-respiratory synchrogram
• dominant Lyapunov exponent
• Electrodermal response
• Emotion recognition
• Eye gaze patterns
• High Order Statistics
• Laguerre expansion
• nonlinear analysis
• Point Process
• PSYCHE platform; Autonomic nervous system monitori
• Textile electrode
• Trispectrum
• Wiener-Volterra Model

This thesis aims at investigating how electrophysiological signals related to the autonomic nervous system (ANS) dynamics could be source of reliable and effective markers for mood state recognition and assessment of emotional responses. In-depth methodological and applicative studies of biosignals such as electrocardiogram, electrodermal response, and respiration activity along with information coming from the eyes (gaze points and pupil size variation) were performed. Supported by the current literature, I found that nonlinear signal processing techniques play a crucial role in understanding the underlying ANS physiology and provide important quantifiers of cardiovascular control dynamics with prognostic value in both healthy subjects and patients.
Two main applicative scenarios were identified: the former includes a group of healthy subjects who was presented with sets of images gathered from the International Affective Picture System hav- ing five levels of arousal and five levels of valence, including both a neutral reference level. The latter was constituted by bipolar patients who were followed for a period of 90 days during which psychophysical evaluations were performed. In both datasets, standard signal processing techniques as well as nonlinear measures have been taken into account to automatically and accurately recognize the elicited levels of arousal and valence and mood states, respectively. A novel probabilistic approach based on the point-process theory was also successfully applied in order to model and characterize the instantaneous ANS nonlinear dynamics in both healthy subjects and bipolar patients. According to the reported evidences on ANS complex behavior, experimental results demonstrate that an accurate characterization of the elicited affective levels and mood states is viable only when non- linear information are retained. Moreover, I demonstrate that the instantaneous ANS assessment is effective in both healthy subjects and patients.
Besides mathematics and signal processing, this thesis also contributes to pragmatic issues such as emotional and mood state mod- eling, elicitation, and noninvasive ANS monitoring. Throughout the dissertation, a critical review on the current state-of-the-art is reported leading to the description of dedicated experimental protocols, reliable mood models, and novel wearable systems able to perform ANS monitoring in a naturalistic environment.