PUB - Publikationen an der Universität Bielefeld

We present a method which removes the need for manual annotation of eye-movement data. Our software produces as output object and subject specific results for various eye tracking parameters in complex 3D scenes. We synchronized a monocular mobile eyetracking system with a VICON motion-capture system. Combining the data of both systems, we calculate and visualize a 3D gaze vector within the VICON coordinate frame of reference. By placing markers on objects and subjects in the scene, we can automatically compute how many times and where fixations occurred. We evaluated our approach by comparing its outcome for a calibration and a grasping task (with three objects: cup, stapler, sphere) against the average results given by the manual annotation. Preliminary data reveals that the program only differs from the average manual annotation results by approximately 3 percent in case of the calibration procedure, where the gaze is subsequently directed towards five different markers on a board, without jumps between them. In case of the more complicated grasping videos the results depend on the object size: for bigger objects (i.e., sphere) the differences in the number of fixations are very small and the cumulative fixaton duration deviates by less than 16 percent (or 950ms). For smaller objects, where there are more saccades towards object boundaries, the differences are bigger. For one reason manual annotation becomes inevitably more subjective; on the other hand both methods analyze the 3D scene from slightly different perspectives (i.e., center of eyeball versus position of scene camera). Although, even then the automatic results come close to those of a manual annotation (the average differences are 984ms and 399ms for the object and hand, respectively) and reflect the fixation distribution when interacting with objects in 3D scenes. Thus, eye-hand coordination experiments with various objects in complex 3D scenes, especially with bigger and moving objects, can now be realized fast and effectively. Our approach allows the recording of eye-, head-, and grasping movements when subjects interact with objects or systems. This allows us to study the relation between gaze and hand movements when people grasp and manipulate objects or indeed free movements in normal gaze behavior. The automatic analysis of gaze and movement data in complex 3D scenes can be applied to a variety of research domains, i.e., Human Computer Interaction, Virtual reality or grasping and gesture research.