Recognising Complex Mental States from Naturalistic Human-Computer Interactions
Access status:
Open Access
Type
ThesisThesis type
Doctor of PhilosophyAuthor/s
Monkaresi, HamedAbstract
New advances in computer vision techniques will revolutionize the way we interact with computers, as they, together with other improvements, will help us build machines that understand us better. The face is the main non-verbal channel for human-human communication and contains ...
See moreNew advances in computer vision techniques will revolutionize the way we interact with computers, as they, together with other improvements, will help us build machines that understand us better. The face is the main non-verbal channel for human-human communication and contains valuable information about emotion, mood, and mental state. Affective computing researchers have investigated widely how facial expressions can be used for automatically recognizing affect and mental states. Nowadays, physiological signals can be measured by video-based techniques, which can also be utilised for emotion detection. Physiological signals, are an important indicator of internal feelings, and are more robust against social masking. This thesis focuses on computer vision techniques to detect facial expression and physiological changes for recognizing non-basic and natural emotions during human-computer interaction. It covers all stages of the research process from data acquisition, integration and application. Most previous studies focused on acquiring data from prototypic basic emotions acted out under laboratory conditions. To evaluate the proposed method under more practical conditions, two different scenarios were used for data collection. In the first scenario, a set of controlled stimulus was used to trigger the user’s emotion. The second scenario aimed at capturing more naturalistic emotions that might occur during a writing activity. In the second scenario, the engagement level of the participants with other affective states was the target of the system. For the first time this thesis explores how video-based physiological measures can be used in affect detection. Video-based measuring of physiological signals is a new technique that needs more improvement to be used in practical applications. A machine learning approach is proposed and evaluated to improve the accuracy of heart rate (HR) measurement using an ordinary camera during a naturalistic interaction with computer.
See less
See moreNew advances in computer vision techniques will revolutionize the way we interact with computers, as they, together with other improvements, will help us build machines that understand us better. The face is the main non-verbal channel for human-human communication and contains valuable information about emotion, mood, and mental state. Affective computing researchers have investigated widely how facial expressions can be used for automatically recognizing affect and mental states. Nowadays, physiological signals can be measured by video-based techniques, which can also be utilised for emotion detection. Physiological signals, are an important indicator of internal feelings, and are more robust against social masking. This thesis focuses on computer vision techniques to detect facial expression and physiological changes for recognizing non-basic and natural emotions during human-computer interaction. It covers all stages of the research process from data acquisition, integration and application. Most previous studies focused on acquiring data from prototypic basic emotions acted out under laboratory conditions. To evaluate the proposed method under more practical conditions, two different scenarios were used for data collection. In the first scenario, a set of controlled stimulus was used to trigger the user’s emotion. The second scenario aimed at capturing more naturalistic emotions that might occur during a writing activity. In the second scenario, the engagement level of the participants with other affective states was the target of the system. For the first time this thesis explores how video-based physiological measures can be used in affect detection. Video-based measuring of physiological signals is a new technique that needs more improvement to be used in practical applications. A machine learning approach is proposed and evaluated to improve the accuracy of heart rate (HR) measurement using an ordinary camera during a naturalistic interaction with computer.
See less
Date
2014-03-27Faculty/School
Faculty of Engineering and Information Technologies, School of Electrical and Information EngineeringAwarding institution
The University of SydneyShare