Measuring gaze and pupil in the real world: object-based attention,3D eye tracking and applications

This dissertation contains studies on visual attention, as measured by gaze orientation, and the use of mobile eye-tracking and pupillometry in applications. It combines the development of methods for mobile eye-tracking (studies II and III) with experimental studies on gaze guidance and pupillary responses in patients (studies IV and VI) and healthy observers (studies I and V). Object based attention / Study I What is the main factor of fixation guidance in natural scenes? Low-level features or objects? We developed a fixation-predicting model, which regards preferred viewing locations (PVL) per object and combines these distributions over the entirety of existing objects in the scene. Object-based fixation predictions for natural scene viewing perform at par with the best early salience model, that are based on low-level features. However, when stimuli are manipulated so that low-level features and objects are dissociated, the greater prediction power of saliency models diminishes. Thus, we dare to claim, that highly developed saliency models implicitly obtain object-hood and that fixation selection is mainly influenced by objects and much less by low-level features. Consequently, attention guidance in natural scenes is object-based. 3D tracking / Study II The second study focussed on improving calibration procedures for eye-in-head positions with a mobile eye-tracker.We used a mobile eye-tracker prototype, the EyeSeeCam with a high video-oculography (VOG) sampling rate and the technical gadget to follow the users gaze direction instantaneously with a rotatable camera. For a better accuracy in eye-positioning, we explored a refinement in the implementation of the eye-in-head calibration that yields a measure for fixation distance, which led to a mobile eye-tracker 3D calibration. Additionally, by developing the analytical mechanics for parametrically reorienting the gaze-centred camera, the 3D calibration could be applied to reliably record gaze-centred videos. Such videos are suitable as stimuli for investigating gaze-behaviour during object manipulation or object recognition in real worlds point-of-view (PoV) perspective. In fact, the 3D calibration produces a higher accuracy in positioning the gaze-centred camera over the whole 3D visual range. Study III, eye-tracking methods With a further development on the EyeSeeCam we achieved to record gaze-in-world data, by superposing eye-in-head and head-in-world coordinates. This novel approach uses a combination of few absolute head-positions extracted manually from the PoV video and of relative head-shifts integrated over angular velocities and translational accelerations, both given by an inertia measurement unit (IMU) synchronized to the VOG data. Gaze-in-world data consist of room-referenced gaze directions and their origins within the environment. They easily allow to assign fixation targets by using a 3D model of the measuring environment – a strong rationalisation regarding fixation analysis. Applications Study III Daylight is an important perceptual factor for visual comfort, but can also create discomfort glare situations during office work, so we developed to measure its behavioural influences. We achieve to compare luminance distributions and fixations in a real-world setting, by also recording indoor luminance variations time-resolved using luminance maps of a scenery spanning over a 3pi sr. Luminance evaluations in the workplace environment yield a well controlled categorisation of different lighting conditions and a localisation as well as a brightness measure of glare sources.We used common tasks like reading, typing on a computer, a phone call and thinking about a subject. The 3D model gives the possibility to test for gaze distribution shifts in the presence of glare patches and for variations between lighting conditions. Here, a low contrast lighting condition with no sun inside and a high contrast lighting condition with direct sunlight inside were compared. When the participants are not engaged in any visually focused task and the presence of the task support is minimal, the dominant view directions are inclined towards the view outside the window under the low contrast lighting conditions, but this tendency is less apparent and sways more towards the inside of the room under the high contrast lighting condition. This result implicates an avoidance of glare sources in gaze behaviour. In a second more extensive series of experiments, the participants’ subjective assessments of the lighting conditions will be included. Thus, the influence of glare can be analysed in more detail and tested whether visual discomfort judgements are correlated in differences in gaze-behaviour. Study IV The advanced eye-tracker calibration found application in several following projects and included in this dissertation is an investigation with patients suffering either from idiopathic Parkinson’s disease or from progressive supranuclear palsy (PSP) syndrome. PSP’s key symptom is the decreased ability to carry out vertical saccades and thus the main diagnostic feature for differentiating between the two forms of Parkinson’s syndrome. By measuring ocular movements during a rapid (< 20s) procedure with a standardized fixation protocol, we could successfully differentiate pre-diagnosed patients between idiopathic Parkinson’s disease and PSP, thus between PSP patients and HCs too. In PSP patients, the EyeSeeCam detected prominent impairment of both saccade velocity and amplitude. Furthermore, we show the benefits of a mobile eye-tracking device for application in clinical practice. Study V Decision-making is one of the basic cognitive processes of human behaviours and thus, also evokes a pupil dilation. Since this dilation reflects a marker for the temporal occurrence of the decision, we wondered whether individuals can read decisions from another’s pupil and thus become a mentalist. For this purpose, a modified version of the rock-paper-scissors childhood game was played with 3 prototypical opponents, while their eyes were video taped. These videos served as stimuli for further persons, who competed in rock-paper-scissors. Our results show, that reading decisions from a competitor’s pupil can be achieved and players can raise their winning probability significantly above chance. This ability does not require training but the instruction, that the time of maximum pupil dilation was indicative of the opponent’s choice. Therefore we conclude, that people could use the pupil to detect cognitive decisions in another individual, if they get explicit knowledge of the pupil’s utility. Study VI For patients with severe motor disabilities, a robust mean of communication is a crucial factor for well-being. Locked-in-Syndrome (LiS) patients suffer from quadriplegia and lack the ability of articulating their voice, though their consciousness is fully intact. While classic and incomplete LiS allows at least voluntary vertical eye movements or blinks to be used for communication, total LiS patients are not able to perform such movements. What remains, are involuntarily evoked muscle reactions, like it is the case with the pupillary response. The pupil dilation reflects enhanced cognitive or emotional processing, which we successfully observed in LiS patients. Furthermore, we created a communication system based on yes-no questions combined with the task of solving arithmetic problems during matching answer intervals, that yet invokes the most solid pupil dilation usable on a trial-by-trial basis for decoding yes or no as answers. Applied to HCs and patients with various severe motor disabilities, we provide the proof of principle that pupil responses allow communication for all tested HCs and 4/7 typical LiS patients. Résumé Together, the methods established within this thesis are promising advances in measuring visual attention allocation with 3D eye-tracking in real world and in the use of pupillometry as on-line measurement of cognitive processes. The two most outstanding findings are the possibility to communicate with complete LiS patients and further a conclusive evidence that objects are the primary unit of fixation selection in natural scenes