Dynamic scanpaths: eye movement analysis methods

An eye movement sequence, or scanpath, during viewing of a stationary stimulus has been described as a set of fixations onto regions-of-interest, ROIs, and the saccades or transitions between them. Such scanpaths have high similarity for the same subject and stimulus both in the spatial loci of the ROIs and their sequence; scanpaths also take place during recollection of a previously viewed stimulus, suggesting that they play a similar role in visual memory and recall (Brandt and Stark, 1997). In a dynamic scenario the features corresponding to ROIs change their absolute and relative positions in time. We introduce a concept of `dynamic scanpath , which utilizes feature-based ROIs, so that a smooth pursuit tracking can be treated as equivalent to a fixation. Strings of the ROIs can be used to represent dynamic scanpaths in time. Quantitative comparisons with string analysis techniques are thus applicable to dynamic as well as static scanpaths. We compared dynamic scanpaths for three subjects (four trials per subject) on eight stimuli, lasting 10 seconds each. The same three-dimensional road scene was presented using three different viewpoint motions (stationary, panning and zooming) and two world models - with and without the cognitive structure. In the latter case, road and grass are replaced by a uniform background and cars by spherical blobs, so that the sequence loses its resemblance to a road scene. We found that dynamic scanpaths exist and can be described similarly to their static counterparts if a smooth pursuit is treated as an extended fixation. Highly different scenes within the same virtual world result in highly different dynamic scanpaths. Viewpoint motion has only slight effect on the scanpaths expressed as strings. Abstraction of visual scene does not critically affect the scanpath. However, there seems to be no common viewing strategy for the abstracted scene, whereas in the non-ed scene there is greater similarity among subjects. The repetitive correlation (same-subject / same-view) was the highest in all cases, suggested the use of a top-down cognitive model and thus provided evidence for the scanpath theory. Statistical significance of the results was evaluated using ANOVA analysis.