Keywords

1 Introduction

Visual fatigue in the VDT, manifested as (1) painful irritation (burning) accompanied by lachrymation, reddening of the eyes and conjunctivitis; (2) double vision; (3) headaches; (4) reduced power of accommodation and convergence; and (5) reduced visual activity, sensitivity to contrast, and speed of perception [1]. It brings a lot of trouble to work and life [2]. Close-up visual tasks for a prolonged time straining the ciliary muscles may cause abnormalities in the accommodative function of the lens,which is known as pseudo myopia and is considered to be a part of refractive myopia. It’s an important cause of VDT visual fatigue [3]. Studies using objective test methods found that although the response to stimuli in patients with visual fatigue was normal, there was a significant minor fluctuation of accommodation [4]. Tosha et al.’s findings showed that patients with visual fatigue tended to have an accommodation delay after gazing at a close-range target for a period of time (90 s or longer), and the accommodation lag that occurred when patients with mild visual fatigue symptoms continued closely gazing remained stable. In patients with severe symptoms, the amount of accommodation lag increased with time [5]. The performance of accommodative dysfunction is poor accommodative facility, and insufficient eye accommodation etc. Accommodative dysfunction can be alleviated by accommodative training, which is derived from a common knowledge that by relaxing the contracted focus-adjustment muscles around the eyeball, known as the ciliary and extraocular muscles, the degree of pseudo myopia can be reduced.

Takada et al. used stereoscopic video clips to train subjects with visual fatigue [3]. The results showed that the visual acuity of the subjects was significantly improved by continuous accommodation training [2]. Sterner et al. studied the effect of flip lens-training on accommodative function, and their results showed that accommodative training significantly increased the accommodative facility and accommodative function of the subjects, and the subjects did not regain any subjective symptoms in the next two years [6]. Our research aims to apply the accommodative training to the head-mounted display virtual reality environment, and explore whether the visual fatigue can be reduced when subjects watching 2D video in the virtual environment.

Human 3D perception is due to the existence of distance between the two eyes, which makes the imaging of object in the retina have a slight difference. This difference is processed through the human visual system to produce depth perception. For the HMD, a pair of optical systems are usually used to create depth perception, which include two small screens to guide the users to receive the left and right eye images with disparity [7]. But the stereoscopic display is somewhat different from human vision. The binocular vision is achieved through the combination of convergence and adjustment mechanisms. In the human visual system, the two mechanisms are tightly coupled because the stimuli that drive them are consistent [8]. But for the head-mounted displays, the virtual image is focused at a fixed depth away from the eyes, while the depth of the virtual objects, and hence the binocular disparity, varies with the content [9,10,11]. When viewing a video, the eyes accommodate to a fixed screen distance while they converge to the simulated distance of the object of interest. For the design of experiment, we change the relative disparity of the human eye fusion by changing the distance between the left and right eye images of the helmet, and generate the depth change of the screen to simulate the adjustment mechanism of the human eye when viewing the near and far objects in the real environment. Since the focal plane of the helmet is fixed, our experiment is essentially to explore whether the physical adjustment and physiological adjustment of the human eye will work under the influence of cognitive adjustment, thereby achieving the expectation of alleviating the dysfunction of accommodation and reducing visual fatigue. We compared the visual fatigue of dynamic disparity groups with the static disparity group.

Takeda et al. studied the characteristics of accommodation evoked by perceived depth sensation. The subjects looked at three different two-dimensional stimuli and two different three-dimensional stimuli. For the two-dimensional stimuli, a manifest accommodation without any accompanying vergence was found because of an apparent depth sensation even though the target distance was kept constant. For the three-dimensional stimuli, larger accommodation and clear vergence were evoked because of binocular disparity and a stronger depth sensation. These results revealed that brain depth perception had an effect on accommodation [12].

Therefore, in order to achieve the effect of relieving visual fatigue of VDT in the virtual reality environment, it is worthwhile to study the influence of creating different depth perceptions through dynamic disparity. We designed and performed an experiment using a combination of subjective and objective assessment to explore the effects of different disparity conditions on visual fatigue. According to the changes of the blinking data, visual fatigue within one hour of the experiment were discussed.

2 Materials

2.1 Hardware System

In order to test whether dynamic disparity can alleviate visual fatigue, the evaluation experiment was carried out. Subjects wore HTC VIVE head-mounted display to view different videos in virtual scenes (Fig. 1). The display has a viewing angle of 110° and a combined resolution of 1200 × 2160 pixels or 1200 × 1080 pixels per eye. The HTC VIVE system also includes a tracking and positioning system with two base stations and an interactive control handle. In order to monitor the eye condition of the subject in real time, the aSee Pro VR eye tracker combined with the HMD was used to obtain the blink data. The computers running the virtual scene and the eye tracker are all above the NVIDIA GeForce GTX 960. The experimental room was managed, with proper illumination conditions, no glare scattered light source and noise isolation. The subjects sat in a comfortable chair to avoid physical fatigue.

Fig. 1.
figure 1

The experimental scene. (a) Illustration of a user watching video through HMD. (b) The schematic of our system.

Full size image

2.2 Software System

Subjects used HMD to view the documentary in a virtual scene created by Unity3d. The scale of the virtual environment and the real environment is consistent. The scene is a simple room of 6 m * 8.5 m * 4 m with a screen in it, as shown in Fig. 2. This is a within-subject experiment. According to the three conditions, it consisted of three groups: a control group with a static disparity in which the images of the left and right eyes overlapped (static group), a group of disparity changing at a constant speed with a screen motion of 0.01 m/s (constant speed group), and a group of disparity changing at a variable speed with jumping 0.05 m per 5 s (non-constant speed group). These two speed settings also ensure that the total motion distances of the screen are the same throughout the experiment.

Fig. 2.
figure 2

The virtual scene, as seen through the HMD.

Full size image

The disparity change of the screen in the human eyes is generated by the horizontal movement of the left and right eye cameras presenting the screen. As shown in Fig. 3(a), the size of the two screens are both 2.76 m * 1.55 m, and the vertical distance of the plane where the screen is located is 3 m from the camera. The field angle is 45°. When the left eye camera is on the right side of the center and the right eye camera is on the left side, it corresponds to the cross disparity, and conversely, the corresponding non-cross disparity. The subjects will have a relatively low visual fatigue when the binocular disparity is around −0.2° to 0.2° [13]. The conventional recommendation for the stable disparity size is generally within 1° [14]. In this case, this experiment ensured that the disparity is within 1°, and the specific range and moving speed were finally determined by a preliminary experiment. For the dynamic disparity groups in the adjustment plane, the center of the left eye image moves from 0.6 cm of the left of the point O to −2 cm of the right of the O (The right eye image moves symmetrically with respect to the left eye image), at this time the screen at the converging surface will move far from near, and the angle of view will remain 45°, as shown in Fig. 3(b).

Fig. 3.
figure 3

(a) Disparity adjustment schematic diagram. (b) The changes of the screen in vergence plane.

Full size image

2.3 Stimuli

The documentaries used in experiment were from the first three episodes of National Treasures, each duration is 1 h. These three videos’ brightness and content are similar. A group watched an episode of video. To rule out the possibility of content affecting visual fatigue, this paper quantitatively evaluated the impact of the selected three videos’ content complexity on blinking [15]. Among the 14 features for texture analysis on the gray level co-occurrence matrix (GLCM), only 4 features are irrelevant [16, 17]. Among these four features, the entropy value is a measure of the amount of information in the image, which indicates the complexity of the texture in the image. Therefore, a Pearson correlation analysis was carried out between the mean entropy value of every 30 s of the videos and the average value of the blink numbers in every 30 s [18].

In the calculation, a 1/12 down sampling of the image is performed. Then it was compressed into 16 gray levels. The results showed that for the static group, the correlation coefficient between the entropy and the blinks is −0.208, which is significant at the level of confidence of 0.95 (p = 0.03 < 0.05). It can be found that the correlation between entropy and blinks is very weak, indicating that the change of entropy of experimental video is not enough as a factor to affect visual fatigue. For the non-constant speed group, the Pearson correlation coefficient is 0.021, and is not significant (p = 0.831 > 0.05). For the constant speed group, the Pearson correlation coefficient is −0.075, and is not significant (p = 0.434 > 0.05). Therefore, in the experimental group, the entropy value change of the video used in the experiment is not a factor that can affect the blink. It could be concluded that the content of the videos we chose didn’t affect the blink results which suggested the visual fatigue state.

3 Methods

3.1 Subjects

17 subjects (10 males and 7 females), most of whom were recruited from Beijing Institute of Technology, participated in the entire experimental process. Before the experiment, they were asked to fill in the basic information and eye disease questionnaire. The age of subjects ranged between 20 and 25. Their myopia was below −2.0D and binocular astigmatism was less than 1D. The subjects had normal stereo vision with no color weakness and color blindness. There was no disease in their eyes. Prior to the experiment, all subjects signed an informed consent form.

3.2 Experimental Procedure

This experiment is a within-subject design. The experiment was conducted during the day and the subjects were asked not to stay up late. As shown in Fig. 4, the experimental process was divided into three stages. Before the experiment, the subjects were first familiar with the experimental instrument and rested for 20 min. They could not watch any electronic screen during the break. Then they filled out the basic information scale and the VFS and SSQ. In the experiment, the subject wore an HMD combined with an eye tracker and sat in a soft chair to watch video for an hour. After the experiment, the subjects filled in the VFS and SSQ again. The above experiment was carried out at the same time for three consecutive days. The first day was the control condition, the second day was the non-constant speed condition, and the third day was the constant speed condition.

Fig. 4.
figure 4

Procedure of experiment. Q: questionnaire, Disparity 1: static group, Disparity 2: constant speed group, Disparity 3: non-constant group.

Full size image

3.3 Measurements

For the assessment of visual fatigue, we adopt a combination of subjective and objective evaluation. The subjective questionnaires were visual fatigue scale (VFS) and simulator sickness questionnaire (SSQ), as it can cause motion sickness in the virtual reality environment [14, 19, 20]. SSQ has 16 items that can test symptoms of nausea, oculomotor and disorientation. VFS consists of 24 items, and the five symptoms that can be tested for visual fatigue are: eye strain, general discomfort, nausea, focusing difficulty, and headache.

The objective measurement was blink characteristics, Blinking is a basic function of the eye and helps to remove corneal and conjunctival irritation. The increased visual load is associated with an increasingly uncomfortable dryness sensation, so the increased blinking frequency is considered evidence of visual fatigue [21,22,23,24]. Eye blinking data has recently been widely used as a feature of visual fatigue [15, 25].

In this experiment, blink information was obtained by eye tracker. The eye images of the subjects were obtained in real time through the aSee Pro VR eye tracker and the blink data in the experiment were processed through the eye tracker supporting software. As shown in Fig. 1(b), the eye tracker was embedded in the HTC VIVE helmet, which illuminated the eye through an infrared LED ring, recorded eye movements in real time through small camera and sent the data back to the computer for processing. It was determined to be blinking when one eye is closed. In order to eliminate the error caused by different states of the same subject under three experimental conditions and difference between different subjects, the method of calculating the blink rate growth radio was adopted. We took the average blink frequency of the first five minutes as a reference, then calculated the growth rate of the nth five-minute’s average blink frequency relative to it.

4 Results

4.1 Results of Subjective Measurements

In order to assess whether the experimental conditions would cause motion sickness and obtain the subjects’ assessment of the visual discomfort, we had the subjects to fill in the SSQ and VFS before and after the experiment. The average changes of the scores before and after the experiment are presented in Fig. 5. The higher the score, the more serious the symptoms. The three factors of the SSQ scale and the total are multiplied by the corresponding weights.

Fig. 5.
figure 5

The mean changes in factor scores. (a) The score of each symptom in SSQ. (N: nausea, O: oculomotor, D: disorientation) (b) The score of each symptom in VFS.

Full size image

As displayed in the figure, the scores of these two scales are both the highest in the static group. Except the focusing difficulty symptom, scores in the constant speed group are lower compared to the non-constant speed group. The differences of the scores between the groups were analyzed by the Friedman method. For SSQ, the differences between the three groups in nausea score (χ^2 = 13.632, P = 0.001 (<0.05)), and total score (χ^2 = 8.291, P = 0.016 (<0.05)) are significant. The differences of oculomotor and disorientation between the three groups are not significant. For the results of VFS, the difference in eye strain score (χ^2 = 10.773, P = 0.005 (<0.05)) is significant, and other differences in general discomfort, nausea, focusing difficulty, and headache are not significant between the three groups.

This result shows that dynamic disparity groups had lighter simulator sickness and visual fatigue symptoms than the static group. The result that eye strain score of VFS is lower in the dynamic disparity groups is significant. The nausea and total scores of SSQ are lower in the dynamic disparity groups and are significant, indicating that instead of causing more motion sickness the dynamic disparity conditions can alleviate it to a certain extent. This may be due to its reduction of visual discomfort. In the dynamic disparity groups, the scores of the above symptoms are the lowest in the condition of constant speed group, indicating that the constant change of disparity can better alleviate motion sickness and visual discomfort. As for the same item—nausea in the two scales, the results’ significance is different. That because their sub-items are different, and in the significant SSQ questionnaire, there are more sub-items.

4.2 Results of Objective Measurements

The growth ratio of average blink frequency per five minutes are presented in Fig. 6. It can be seen that in the first 15 min, the blink frequency ratio of the three groups decreased slightly; In the last ten minutes, the highest value was reached at 50 min, and decreased at 55 min; The blink frequency ratio was in a stable fluctuation in the static group and same in the constant speed group at 20–45 min. According to this, it can be divided into three stages: 10–15, 20–40, 45–55 min.

Fig. 6.
figure 6

The growth ratio of average blink frequency per five minutes to the initial five minutes.

Full size image

The trend of the constant speed group and the static group is similar. While the variable speed group goes up in the middle stage. Its blink frequency ratio increased at 15–35 min and decreased at 40 min, after which it increased in 45 min. A two-factor repeated measures analysis of variance was used to analyze the effects of different disparity condition over time on the blink frequency ratio of the subjects. It aimed to determine whether there was a significant difference in the blink frequency ratio between the three different disparity condition groups in the significance level α = 0.05. The statistics are performed using SPSS 22.0. The data contains two dependent variables: different disparity conditions (v) and time periods (time).

In the 10–15 min, for the interaction term v * time (\( \upchi^{2} = 3.056 \), P = 0.217 (>0.05)) and v (\( \upchi^{2} = 3.200 \), P = 0.202 (>0.05)), the dependent variable satisfies the spherical hypothesis. The effect test in the subject indicates that the interaction term is not statistically significant (F(2, 32) = 0.167, P = 0.847 (>0.05)); v has no statistically significant effect on blinking (F(2, 32) = 2.962, P = 0.066 (>0.05)); time has no statistically significant effect on blinking (F(1, 16) = 0.232, P = 0.636 (>0.05)). Thus, the blink frequency ratio at this stage is not significantly different between the three groups.

In the 20–40 min, for the interaction term v * time (\( \upchi^{2} = 67.686 \), P = 0.001 (<0.05)), the dependent variable does not satisfy the spherical hypothesis, and is corrected using Greenhouse-Geisser correction; v satisfies the spherical hypothesis (\( \upchi^{2} = 3.194 \), P = 0.203 (>0.05)); time (\( \upchi^{2} = 19.187 \), P = 0.024 (<0.05)) is corrected using Greenhouse-Geisser. The effect test within the subject indicates that the interaction term is not statistically significant (F(4.203, 67.246) = 0.923, P = 0.460 (>0.05)); v has statistically significant effect on blink (F(2, 32) = 9.452, P = 0.001 (<0.05)); time has no statistically significant effect on blinking (F(2.673, 42.776) = 0.214, P = 0.866 (>0.05)). Regarding the results of the pairwise comparison, the difference between the static group and the non-constant speed group is significant (P = 0.03 < 0.05), and the average difference of the blink frequency ratio is −0.430 ± 0.108; There is no significant difference between the static group and the constant speed group (P = 0.074 > 0.05). There is no significant difference between the non-constant speed group and the constant speed group (P = 0.129 > 0.05). That is, the blink ratio of the non-constant speed group is significantly higher than that of the static group, and the difference between the other groups is not statistically significant.

In the 45–55 min, the interaction term v * time satisfies the spherical hypothesis (\( \upchi^{2} = 13.184 \), P = 0.157 (>0.05)); v satisfies the spherical hypothesis (\( \upchi^{2} = 2.462 \), P = 0.292 (>0.05)); time (\( \upchi^{2} = 10.463 \), P = 0.005 (<0.05)) is corrected using Greenhouse-Geisser. The effect test in the subject indicates that the interaction term is not statistically significant (F(4, 64) = 0.115, P = 0.977 (>0.05)); v has statistically significant effect on blink (F(2, 32) = 4.555, P = 0.018 (<0.05)); time has statistically significant effect on blink (F (1.331, 21.302) = 14.081, P < 0.05). Regarding the results of comparing v with each other, the difference between the static group and the non-constant speed group is significant (P = 0.004 < 0.05), and the average difference in the blink frequency ratio is −0.438 ± 1.32. There is no significant difference between the static group and the constant speed group (P = 0.148 > 0.05). There is no significant difference between the non-constant speed group and the constant speed group (P = 0.176 > 0.05). Regarding the results of comparing time with each other, the 45 min’ blink frequency ratio with 50 min’ blink frequency ratio and it with 55 min’ blink frequency ratio are significantly different. The average difference is −0.274 ± 0.061 and −0.152 ± 0.028, respectively. There is no significant difference in the blink frequency ratio between 50 min and 55 min (P = 0.175). That is, at the last stage, the blink frequency ratio of the non-constant speed group is significantly higher than that of the static group, and the difference between the other groups is not statistically significant. And at the 50 min, the blink frequency ratio increases to the top in all groups.

This result shows that the growth trend of the blink frequency ratio in order of high-to-low is non-constant speed group, constant speed group and static group. In 20–40 min and 45–55 min, the difference between the static group and the on-constant speed group is significant. There is no significant difference between the constant speed group with the other two groups. This indicates that the objective visual fatigue of the subjects is the lowest in the static group and higher in the dynamic disparity groups. Among the dynamic disparity groups, the constant speed group had lower visual fatigue than the non-constant speed group. The blinks of the three groups reached their peak in 45–50 min.

5 Discussion

There is a distinction between the subjective and objective measurements. For the objectively measured results, it showed a decrease in performance of the human vision system, which refers to visual fatigue. Whereas the visual discomfort is its subjective counterpart [26]. This experiment’s results show that the subjects’ visual discomfort reduced in the dynamic disparity condition, especially in the constant speed condition. While the visual fatigue is higher in the dynamic disparity groups, especially in the variable speed condition. Koulieris et al. as well as other work by Konrad et al showed that manipulating disparity alone could not cause a significant change of eye accommodation [27, 28]. That could be the reason that the visual fatigue can’t be reduced by dynamic disparity. As for the visual discomfort, maybe there was a delay in the feeling of the subjects, or the mechanism of the two is different.

According to the blink data, the ratio of blinking frequency in the static group and the constant speed group can be divided into three stages: declining (0–15 min), smooth fluctuation (15–45 min) and rising (45–50 min). While the non-constant speed group can be divided into four stages: declining (0–15 min), rising (15–30 min), declining (30–45 min) and rising (45–50 min). And they all started to grow rapidly in 45–50 min until the peak, which dropped in 50–55 min. These phenomena may be due to the fact that the subject had an adaptation process to the scene at the beginning. Since the variable-speed disparity can cause visual fatigue more quickly, the blink frequency started to increase first. As the fatigue of the subject increased, the duration of the blink increased, so the blink frequency decreased a little, but it was still higher than the base value [29]. So, at 40 min of the variable speed group, and 55 min of all three groups, there is a drop. The most notable is that at 50 min, the blink frequency ratio of each group reaches the highest value. The video content was not the end there, and had no difference from the content at other time periods. This shows that there is a significant increase of fatigue in 45–50 min when watching 2D video. The study of Guo also indicated that during the 60 min’ task in the VR, the visual fatigue increased severe quickly in the last 20 min [30].

6 Conclusion

This study aimed to explore whether dynamic disparity can alleviate the visual fatigue caused by watching video for long period in head-mounted displays. We designed three sets of controlled trials, in which disparity was used as a control variable. The trials were composed of static group, constant speed group and non-constant speed group. Subjective and objective data were combined to evaluate the visual fatigue.

For the results of the subjective questionnaire, it indicated that the subjects felt lower visual discomfort and motion sickness under dynamic disparity conditions. In the dynamic disparity groups, the constant speed condition made the subjects feel the lowest degree of visual discomfort and motion sickness. For the results of objective measurement, the blink data showed the fatigue changes of subjects who watched 2D video for 55 min on an HMD. It can be seen that the variation tendency of the blink ratio in the static group and the constant speed group is similar, and the blink radio in the constant speed group is higher, but there is no significant difference. There is a significant difference between the non-constant speed group and the static group in the 20–40 min and 45–55 min, and the non-constant speed group has a higher blink ratio. The conclusions reached are inconsistent with the subjective conclusions, it seems that the dynamic disparity condition did not reduce the objective fatigue of the subjects, and in the variable speed group, the visual fatigue increased. According to the analysis of the results, we came to the conclusion that dynamic disparity achieved by the movement of left and right eye images in the HMD can’t effectively alleviate visual fatigue.

There are some limitations in the experiment should be pointed out. In the design of the experiment, the order of the three groups was fixed, although the effect of video content on blinking was excluded. The experiment was carried out according to the static group, the non-constant speed group, and the constant speed disparity group. In the measurement of objective data, only the blink frequency was tested. It can be combined with other characteristics such as blink duration to measure the visual fatigue of the subject for the future work. Based on the results, in order to achieve the goal of relieving the visual fatigue of viewing 2D screens for a long time in a virtual reality environment by accommodative training, improvements in helmet hardware are required.