In this paper, we propose a direction instruction method using active patterns that is both easy to understand and safe in augmented reality fashion. An active pattern is a set of moving virtual objects dynamically arranged in a landscape. We discuss the factors that make active patterns easy to understand and safe. We have developed a preliminary system for presenting active patterns and included the gaze accumulation counter inside the system so that the gazing time of watching the active patterns could be counted.
Augmented Reality is useful for walking pedestrians on
streets [
Direction instruction using an HMD may superimpose the route to go on the scenery. Because the instruction is presented directly in the user’s field of view, the direction to go should be intuitive and easy to understand. The user can recognize the direction to go while always looking forward.
When using augmented reality technology to guide pedestrians to their destinations, the problem is how to superimpose the virtual objects that guide them to their destinations. AR-based direction presentation should fulifll the visibility on display. As the comprehensibility, it should have a good shape and/or motion to indicate the direction to go. In addition to visibility and comprehensibility, it is also important to keep the safety on a walking task. A user should have a clear field of view while walking. This implies that any inserted virtual objects in AR fashion should not interfere with safety.
In this paper, we propose a direction instruction method using active patterns that is both easy to understand and safe in augmented reality fashion. An active pattern is a set of moving virtual objects dynamically arranged in a landscape (Figure 1).
Since the good balance of visibility, comprehensibility, and safety on the AR-based direction presentation is crucial, we investigate the performance of the active patterns by changing the shape, size, height, and transparency of the virtual objects. We have built a preliminary system for evaluation on gaze attraction for checking the safety. We also conduct subjective evaluation by questionnaires.
and it might hide certain size of the field of view of the user.
Performance of AR-based direction display methods
on of-road situation has been studied [
Safety is a critical issue to design the navigation and
to analyze the performance on direction presentation in
AR fashion. It is demanded to focus the safety evaluation
on developing new AR-based navigation approaches [
Our active pattern to present direction on an HMD is composed by a set of virtual objects that are moving along the way to go. We adopt an optical see-through glass as to build the preliminary system for evaluation (Figure 1).
Since the purpose of presenting the direction is to let users understand the direction to go, active pattern should be visually simple and small not to interfere with the the field of view of the user. We have prepared three simple shapes as candidates for the virtual objects (Figure 2). On a see-through HMD, the objects are arranged to follow the planned path. Three snapshots of a user’s view with the diferent virtual object shape at ground level are shown in Figure 3.
Some variation of ball-shape active pattern is shown in Figure 4. Transparency is controlled so that users can see the the regions behind the virtual objects.
We have developed a preliminary AR system on a Magic Leap 1 (Figure 1). The experiment was conducted in a corridor where the planned path is set to turn right at 15 meters away (Figure 5).
All the subjects are familiar with the corridor and with AR experiences in their twenties.
We prepared eight subjects for shape and height evaluation of the virtual objects. Each subject had a calibration process and practice time before the experiment. The order of the variations is changed for subjects so as to eliminate the order efect in total.
There five kinds of trials on this experiment. The first one (static/ground) is the ball shape virtual object with static placement. The objects are at the ground level. For the second to the fourth, the virtual objects of moving along the planned path is rendered at the ground level with the shape of ball, cube and ellipsoid respectively. The last fifth is same as the third except for the height at knee level. The size of the virtual objects are set to 10 cm and transparency level is 0% for all the five trials.
The subjects mark scores with a seven-level Likert scale for four questions after each experience of trials. The average and the standard deviation are shown in Figure 6 - 9. A larger score indicates the afirmative evaluation.
Q1 (Figure 6) corresponds to the visibility and Q2 (Figure 7) corresponds to the comprehensibility. The scores are high with small standard deviation regardless of shape and height if only they are moving. Q3 (Figure 8) and Q4 (Figure 9) corresponds to the safety evaluation. We think if the score of Q3 is high, it means it looks natural, so it may indicate that the score of Q4 becomes low because they do not need to watch the virtual objects for a long time. This also implies the safety could be in good level. As the Q4 score are rather high around 5.0, we have conducted further investigation with gaze analysis in the next experiment.
On the second experiment, we have developed a gaze accumulation counter on our preliminary system. We have utilized the gaze tracking function of Magic Leap 1 and set nine by nine bins to cover the field of view of the user (Figure 10). The bin matrix is fixed to the head position of the user, and it is not moved against the head rotation. It is because the virtual objects are attached to the world and the bin matrix should be relatively fixed against the virtual object coordinate.
We prepared nine subjects for size and transparency evaluation of the virtual objects. The basic procedure is same as the first experiment.
There five kinds of trials on this experiment. For the all five trials, the shape is ball and the objects are placed at ground level and move to follow the planned path. The ifrst three trials changes the size at 20cm, 10cm, and 5cm respectively. As for the fourth and the fifth trials, the transparency level is changed to 50% and 75%.
The subject evaluations are conducted by setting the four questions after each experience of trials. The average and the standard deviation are shown in Figure 11 - 14. A larger score indicates the afirmative evaluation.
From the results, Q1 (visibility) and Q2 (comprehensibility) becomes lower when the transparency level is up. We expected the trials with higher transparency may get higher score in Q3 and lower in Q4, but actually the subjects did not respond as we expected.
Note that we have not shown the results of T-test intentionally as the numbers of the subjects are less than ten and we think the application of T-test is not appropriate on the second experiment, the lower half of the bin matrix corresponds to the area of presenting the virtual objects.
Even with the high scores of Q4 (Figure 14), we can say that actually the subjects spent little time on checking the virtual objects. This means the current implementation satisfies the safety property to some extent.
for Figure 6 - 9 and Figure 11 - 14.
The results of the gaze accumulation counter during In this paper, we proposed a direction instruction method the second experiment are shown in Figure 15 and Fig- using active patterns that is both easy to understand ure 16. Figure 15 corresponds to the first three trials of the and safe in augmented reality fashion with a optical seesecond experiment. Figure 16 corresponds to the second, through HMD. We have developed a preliminary system the fourth, and the fifth trials of the second experiment. for presenting active patterns and included the gaze acThe darkest green color bin indicates the 0.0 time unit of cumulation counter inside the system so that the gazing gaze accumulation and the brightest color bin indicates time of watching the active patterns could be counted. the maximum amount of time unit through the second We have conducted subjective evaluation experiments experiment in the average of all the nine subjects’ trials. and discussed the factors that make active patterns easy time length per time unit is determined to normalize the to understand and safe. length diference between the five trials.
Since the virtual objects are placed on the ground level A part of this work is supported by KAKENHI 21H03476.