Estimating the Gaze Point of a Student in a Driving Simulator


Fikkert, F.W. (2005) Estimating the Gaze Point of a Student in a Driving Simulator. [Masters Thesis]

[img] PDF
Restricted to Repository staff only

Abstract:This thesis describes the research related the “Behaviour Modelling” project at Green Dino, The Netherlands. The goal is to estimate a student’s body pose and focus of attention in “De Nederlandse Rijsimulator” (DNR), which is a fully automated driving simulator. Students follow a curriculum as a preparation on the practical part of their driving course. A virtual driving instructor (VDI) provides feedback during lessons but lacks information regarding student body pose and gaze behaviour. Direct feedback to the student thereon cannot be given. The goal is to indirectly enhance the student’s learning experience by providing the VDI with an estimated student focus of attention and body pose during lessons.

Numerous sensor types were compared. A stereo vision off-the-shelf camera set-up was chosen, observing the student from two views in the DNR. A top down camera provides images in which the student is visible in full, enabling body pose estimation. An en face camera provides full facial images. Analysing those image leads to a gaze direction estimate. Correlation between data obtained from both cameras provides the head’s location. In turn combining that with the gaze direction, the focus of attention is found. A gaze direction is indicated by head orientations. Techniques for body pose and gaze direction estimation are described, each in a literature overview.

A hierarchical, feature-based approach was chosen for both body pose and gaze direction estimation problems. Skin colour detection provides head and hand positions. Using this approach, the face is found in both cameras’ images. Facial features are found as holes in the facial skin patch. The eyes, ears, and mouth were detected in this manner. A feature constellation describes these detected facial features in terms of layout. Such a constellation represents a gaze orientation. A pool of predefined gaze orientations is matched with the currently found constellation, providing accurate estimations of a student’s current focus of attention.
Item Type:Masters Thesis
Electrical Engineering, Mathematics and Computer Science (EEMCS)
Research Group:
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page

Metis ID: 229434