Figure 1.
Subjects performed a visual search task within a virtual-reality three-room apartment.
A. Subjects wore a V8 head mounted display (HMD) equipped with an ASL eye tracker, while the head and hand were tracked using the PhaseSpace motion tracking system. The subject touched virtual objects in the apartment using the gloved hand when they were located. B. Overhead view of the virtual apartment showing the three rooms: bathroom, bedroom, and kitchen. The 3 objects specified for repeated search are indicated by purple circles. C. Two example views, recorded while a human subject was exploring the apartment wearing the HMD. Targets were specified by words at the bottom of the screen (e.g., “Coffee Maker”) and the subject had to locate and touch that object (e.g., the orange coffee maker). Gaze position, head orientation and location were recorded for later analysis.
Figure 2.
The distribution of gaze locations in the apartment on Day 1 reflect an inhomogeneous sampling of space.
A. Heat map of gaze locations in the XZ plane, collapsed over vertical height, Y, within the apartment. B. Distribution of fixations in the XZ plane, ordered by frequency and normalized. C. Histogram of gaze locations on the Y axis, vertical height, normalized. Presented data are grouped over all six subjects for Day 1.
Figure 3.
The number of fixations required to locate a search target decreases over repeated Search Episodes.
Visual search performance is quantified as the number of fixations allocated in space between when the search target appeared in subject's field of view to the time the target was fixated upon. Data are for the three objects that were repeatedly searched for, averaged first over objects and then subjects, for day 1 and day 2. Error bars represent 95% confidence intervals, bootstrapped, between subjects. The curves show an exponential fit over the two days.
Figure 4.
The efficiency with which subjects plan their path to search targets modestly increases over repeated Search Episodes.
Path efficiency (the shortest possible path between subject and search target divided by the actual path taken by the subject to search target, see methods) is plotted as a function of Search Episode. The path is calculated between trial start to the time the target was fixated upon. Data are for the three objects that were repeatedly searched for, averaged over objects and subjects, for day 1 and day 2. Error bars represent 95% confidence intervals, bootstrapped, between subjects. The curves show an exponential fit over the two days.
Figure 5.
The number of incidental fixations to objects that have yet to be searched for does not correlate with number of fixations required to locate the object on 1st Search Episode.
Incidental fixations (x-axis) are considered incidental if the fixation was made to a non-target object before that object has ever been identified as a search target. The number of fixations required to locate the object once it has become a search target for the first time (1st Search Episode) are presented on the y-axis. Each object contributes 1 data point, and identical points are not obvious on the scatterplot. Marginal histograms are therefore presented to the right and above the scatterplot, and distribution means are indicated by the thin lines. SE = Search Episode.
Figure 6.
The three objects that were searched for repeatedly and their color change on day 3 of the experiment.
From left to right: the coffeemaker, bedstand, and kettle. Top row presents the object as it was on day 1, day 2 and day 3 before change, bottom row presents the objects after the day 3 change.
Figure 7.
The probability of fixating an object increases for objects that have changed color, but not for those that have remained unchanged.
Both panels present p(fixation|inFOV), the probability of fixating an object given that it is in the field of view and not a current target of a search, for Day 1, Day 2, Day 3 before the color change and Day 3 after the color change, averaged over objects and subjects. A. Mean p(fixation|inFOV) for the three objects that changed color. B. Mean p(fixation|inFOV) for the remaining unchanged objects. Error bars are standard error between subjects. FOV = field of view. * p>0.05, ** p>0.001, corrected for multiple comparisons.
Figure 8.
Bottom up saliency does not change as a result of object color change.
Left: Image of the coffee maker and corresponding saliency map per code by Itti and Koch [22]. Right: Image of the coffee maker after the color change, and the corresponding saliency map. Saliency values were computed within the red rectangle.