![](bg4.png)
points that randomly appeared one after another on
the screen. The experiments consisted of three stages,
involving the use of the eye-gaze interface (a) with-
out the guides, (b) with the recognition method Guide
1, and (c) with the recognition method Guide 2. The
subjects were randomly ordered in order to avoid the
order effect in these stages. Semi-dynamic calibration
was used if necessary.
5 EXPERIMENTAL RESULTS
Table 1 shows the pointing accuracy for all subjects
in the use of the eye-gaze interface. Each value de-
notes the average accuracy over the 25 target points.
The pointing accuracy for each target point was calcu-
lated from image data of 30 frames that correspond to
1 s. The accuracy varies with the users and the meth-
ods. The results suggest that the pointing accuracy
of the eye-gaze system using our recognition meth-
ods is comparable to that without the methods. We
think that because semi-dynamic calibration worked
effectively to update the mapping function, the three
conditions had no significant effect on improving the
degree of accuracy.
Table 2 shows the time required for each trial.
As shown in the table, the average time required for
Guide 2 is 270.8 s, in contrast to 427.4 s without the
guides. The number of times the semi-dynamic cal-
ibration was activated in each trial is also shown in
Table 3. The shorter times needed suggest that the
guide systems are useful in recognizing the position
of the user’s head and reducing the frequency of situ-
ations where the mouse pointer on the screen does not
follow the user’s gaze.
We performed usability evaluation at the end of
the experiment for each subject. Table 4 shows the
questionnaire items for evaluation. The subjects were
asked questions related to ease of use, accuracy, and
comfort of the system for each of the three conditions:
Without Guides, Guide 1, and Guide 2. Items regard-
ing the recognition of the home position and appropri-
ate adjustment of the head position were added to the
questionnaires for Guide 1 and Guide 2. Because a
polarizing film is attached to the front of the goggles,
as shown in Figure 4(b), an item regarding visibil-
ity was added to the questionnaire for Guide 2. The
final item was the subjects’ overall impression of the
usefulness of the guides.
Table 5 shows the results of the questionnaire.
Items (1) to (14) were evaluated on a grading scale
ranging from 1 (poor) to 5 (good). Item (15) was
evaluated according to the number of subjects who
preferred the method. As indicated in item (15), all
Table 1: Results of pointing accuracy.
Accuracy [pixels]
Subjects Without Guides Guide 1 Guide 2
Subject 1 34.6 44.5 44.1
Subject 2 57.2 65.3 56.3
Subject 3 47.8 36.5 26.1
Subject 4 61.3 84.7 80.7
Subject 5 75.5 53.6 55.3
Average 55.3 56.9 52.5
Table 2: Time required in each trial.
Time [s]
Subjects Without Guides Guide 1 Guide 2
Subject 1 374.9 248.6 189.4
Subject 2 409.1 570.3 307.9
Subject 3 674.0 161.1 258.3
Subject 4 312.1 266.1 252.1
Subject 5 366.9 226.3 346.3
Average 427.4 294.5 270.8
Table 3: The number of uses of semi-dynamic calibration
in each trial.
Number [times]
Subjects Without Guides Guide 1 Guide 2
Subject 1 11 5 3
Subject 2 5 7 6
Subject 3 21 1 4
Subject 4 8 7 5
Subject 5 6 3 3
Average 10.2 4.6 4.2
subjects preferred the system with guides. Among
the three conditions, Guides 1 and 2 recorded higher
scores on the three common items regarding ease of
use, accuracy, and comfort. A comparison of the
scores for items (7), (8), (12), and (13) shows that
Guide 2 is superior to Guide 1 in terms of recog-
nizing the home position and appropriately adjusting
the head position. In a comment form provided in
the questionnaire, a subject pointed out that Guide 2
was more suitable for shortsighted people than Guide
1. Another subject pointed out that the polarizing
film attached to the goggles helped improve visibility.
This suggests that the film eliminated part of the light
from the display and made it easier for the subject to
see.
6 CONCLUSIONS
In this paper, we proposed two methods for recog-
nizing the user’s head position using polarizing films
SIGMAP2014-InternationalConferenceonSignalProcessingandMultimediaApplications
278