Twenty healthy participants without ocular diseases, except for refractive errors (seven males and 13 females), seven protanope participants (seven males), and five deuteranope participants (four males and one female) were included. Their mean age (SD) was 21.0±1.0 years, 46,7±16.1 years, and 56.6±6.9 years, respectively. All participants had corrected visual acuity of >20/20. The mean spherical refractive errors for the normal trichromats, protanope, and deuteranope participants were -2.45±2.66 (-0.25 to -6.00) diopter (D) -2.81±2.39 (-0.25 to -5.50) D and -2.45±2.91 (-1.00 to -6.00) D, respectively, and the mean astigmatism results were -0.72±0.63 D, -0.56±0.81 D, and -0.25±0.27 D, respectively. Refractive errors were fully corrected before the study was performed, and participants with astigmatism greater than -2.00 D were excluded. Congenital color vision defect in protanope and deuteranope participants were diagnosed using an anomaloscope. A cone contrast test was performed on the ColorDX CCT-HD (Konan Medical, Inc., CA, US) to account for the effect of the acquired color vision defect due to yellowing of the crystalline lens. Participants were excluded if their S-cone contrast sensitivity score was less than 75, which is diagnostic of color vision defects (details of each participant’s cone contrast score are shown in S2 Fig). Individuals with a history of ophthalmic surgery, retinal or optic nerve disease, or systemic diseases were excluded. This study was conducted in accordance with the principles of the Declaration of Helsinki. The procedures used were approved by the Ethics Committee of the School of Allied Health Sciences of Kitasato University (2019-018B). Written informed consent was obtained from all participants. The study was conducted in accordance with the guidelines and regulations related to informed consent. In addition, we obtained informed consent from the participants to submit the research results to open-access publications, which may lead to personal identification.

All samples used in this study were made of 100% polyester fabric manufactured by the Sakai Ovex Corporation (Fukui, Japan). A polyvinyl chloride pipe 70 mm in height and 25 mm in diameter was covered by the samples. General and fluorescent colors within the chromaticity coordinate standards for safety colors of ISO 3864–4:2011(en) were selected to undergo visibility evaluation [9]. The colors used were red (x = 0.537, y = 0.366, β = 0.17), yellow (x = 0.445, y = 0.497, β = 0.49), yellow-red (x = 0.500, y = 0.420, β = 0.32), green (x = 0.248, y = 0.423, β = 0.22), blue (x = 0.195, y = 0.219, β = 0. 15), red-purple (x = 0.343, y = 0.247, β = 0.12), white (x = 0.310, y = 0.325, β = 0.64), black (x = 0.314, y = 0.326, β = 0.02), fluorescent yellow (x = 0.375, y = 0.532, β = 1.00), and fluorescent orange (x = 0.553, y = 0.415, β = 0.69). Each color was combined with black and white to form a striped pattern. There were 16 different color combinations: black/florescent yellow stripe, black/white stripe, black/red-purple stripe, black/blue stripe, black/green stripe, black/yellow stripe, black/yellow-red stripe, black/red stripe, white/red-purple stripe, white/blue stripe, white/green stripe, white/yellow stripe, white/yellow-red stripe, white/red stripe, plain fluorescent yellow, and plain fluorescent orange (Fig 1). The stripe pattern was set to 6.0 cycles/degree (c/deg), which is the spatial frequency band of the peak contrast sensitivity, and the ratio of the stripes of the two colors was set to 50%. Sixteen different combinations of samples were compared with 120 patterns. The luminance values of each of the 16 samples were measured using a spot-type single-lens reflex digital luminance meter LS-100 (Konica Minolta, Tokyo, Japan) in a research environment that simulated sunset. Three measurements were taken at a distance of 5.0 m from the sample, each at a measurement angle of 1°, and the average value of each measurement was calculated (Table 1).

The participants were surrounded by a black curtain (approximately 2.5 m) to prevent uneven luminance distribution in the visual field. To reproduce the brightness at sunset in a dark room, a light source was placed in front of the sample so that it was not visible to the examinee, and the visual target illuminance was set to 300 lx (Figs 2 and 3). SFX-502 (Panasonic, Osaka, Japan), a warm white light-emitting diode with a color temperature of approximately 2700 K, was used as the light source. To fix the spatial frequency of the samples, two different samples were placed 5 m away from the line of sight, and two light sources were placed in front of the samples. These simulations of the lighting environment at sunset are based on measured luminance and illuminance color temperatures during actual sunset hours, averaged over weather and interday variations.

First, the refractive error was fully corrected in a light room, and basic data such as the presence or absence of the participant’s medical history, were obtained. Next, we evaluated the visibility of the samples in a simulated sunset environment. For the visibility evaluation, participants were seated, a liquid crystal shutter (Koyo Corporation, Singapore) was placed in front of their eyes, and a shading cylinder with a 4° field of view was placed behind it (Fig 2). By applying this liquid-crystal shutter voltage, the shutter can be opened and closed to instantly change the view from invisible to visible. When the shutter was open, the parallel transmittance was 74%, and when it was closed, the parallel transmittance was 5% (Fig 4). The liquid crystal shutter was connected to a PC and software with a control signal generator AWG10K (Elmos Co., Ltd., Aichi, Japan) and a semiconductor relay AC100SSR (Asakusa Giken Co., Ltd., Tokyo, Japan). The signals were controlled using the dedicated Waveform Editing software (version 1.1, Elmos Co., Ltd.). The open/close function of the shutter was used to show the samples for 0.5 s. This time was based on “the time it takes for a car traveling at 50 km/h to recognize a person 50 m ahead, consider the braking distance of the vehicle, and stop approximately 10 m in front of the person [10]”.

Two combinations of the left and right samples were randomly presented 5 m in front of the participant to evaluate visibility. The left and right samples were evaluated by circling which sample was brighter or more conspicuous as A and B, as shown in Fig 5. Only brightness sensitivity was evaluated in participants with normal trichromats, and in participants with protanopes and deuteranopes, brightness and conspicuity sensitivity were evaluated. Brightness and conspicuousness sensitivity measurements were evaluated separately in the evaluation experiments. Comparing the two samples as A and B, A’s (or B’s) degree of brightness was answered with “very bright” or “bright.” When the left and right samples appeared to be equally bright, or when it was impossible to judge which was brighter, the participants were asked to select “neither.” The conspicuousness of the samples was also evaluated in the same way. The time available to view the two samples was approximately 0.5 s when the liquid crystal shutter was open. All participants were given breaks to ensure sufficient time to fill out the evaluation sheet and avoid fatigue.

Statistical analysis was conducted using Excel 2016 (Microsoft Co., Albuquerque, NM, USA) and Bell Curve for Excel (Social Research Information, Inc., Tokyo, Japan). The analysis was performed using Sheffé’s (Nakaya) paired comparison method. This is a grading method in which any two objects are taken and compared on a one-to-one basis, and the results of all comparisons are evaluated collectively. Analysis of variance (ANOVA) and the average degree of preference corresponding to the rated value of each sample were obtained. Based on the results of the analysis of variance, it was determined that there is a significant difference between samples when the p-value of the main effect is less than 0.05. In this case, a significant difference existed if the difference in the average degree of preference between samples was greater than the yardstick value (Yφ). The yardstick value (Yφ) was obtained using Eq (1).

φ is the significance level to be tested and 0.05 is used. t and n are the numbers of samples and participants to be evaluated, respectively. f_e and V_e are the degrees of freedom and variance of the residuals, respectively, obtained from the analysis of the variance table. q_φ(t, f_e) was also studentized at a significance level of 0.05. For all tests, a P value < 0.05 was considered statistically significant.

Table 2 shows the analysis of variance table for normal trichromats, and protanope and deuteranope using the Scheffé method. The F-test results showed significant differences in the main effects for all groups. Since significant differences were found between normal trichromats and protanope and deuteranope, the yardstick values (Y) obtained using q-values from the studentized range with a 5% level of significance for the degrees of freedom of each error, and the average degree of preference and confidence interval for each sample are listed in Table 3. Normal trichromats had the highest average degree of preference for fluorescent orange at 1.08, with a difference of 0.05 from fluorescent yellow, which is less than the yardstick value of 0.32, and is therefore equally visible. All samples with a lower average degree of preference than fluorescent yellow were significantly different (p < 0.05) from the average degree of preference for fluorescent orange, as they differed by more than 0.32. Protanope had the highest average degree of preference for fluorescent yellow at 1.20, with a difference of 0.09 from the white/yellow, which is less than the yardstick value of 0.47 and therefore of equal visibility. All samples with a lower average degree of preference than the white/yellow were significantly different (p < 0.05) from the fluorescent yellow average degree of preference, as they differed by more than 0.47. The deuteranope had the highest average degree of preference for fluorescent yellow at 1.17, with a difference of 0.22 from white/yellow, which is less than the yardstick value of 0.48 and therefore of equal visibility. All samples with a lower average degree of preference than white/yellow were significantly different (p < 0.05) from the fluorescent yellow average degree of preference, as they differed by more than 0.48. Fig 6 presents a comparison of sample results for all three of these groups with a standardized normal distribution. The distribution is relatively similar for normal trichromats and deuteranope, but the distribution is off for the protanope, especially the low average degree of preference for fluorescent orange.

Table 4 shows the analysis of variance table for protanopes and deuteranopes using the Scheffé method. The F-test results showed significant differences in the main effects for all groups. Since significant differences were found between the rotanope and deuteranope, the yardstick values (Y) were obtained using q-values from the Studentized range with a 5% level of significance for the degrees of freedom of each error; the average degree of preference and confidence interval for each sample are presented in Table 5. Protanope had the highest average degree of preference for black/fluorescent yellow at 1.03, with a difference of 0.14 from the black/yellow, 0.32 from the black/white, 0.44 from the black/yellow-red, and 0.49 from white/red-purple, which is less than the yardstick value of 0.51 and therefore equally visible. All samples with a lower average degree of preference than white/red-purple were significantly different (p < 0.05) from the black/fluorescent yellow average degree of preference, as they differed by more than 0.51. Deuteranope had the highest average degree of preference for black/fluorescent yellow at 1.08, with a difference of 0.13 from black/yellow, 0.22 from the black/white, 0.34 from the black/yellow-red, 0.50 from the white/red-purple, which is less than the yardstick value of 0.50, and therefore equally visible. All samples with a lower average degree of preference than white/red-purple were significantly different (p < 0.05) from the black/fluorescent yellow average degree of preference, as they differed by more than 0.50. Fig 7 compares the sample results for all three groups with a standardized normal distribution. The protanopes and deuteranopes showed similar normal distributions in almost all samples. We analyzed the correlation between brightness and conspicuousness sensitivity, but there was no correlation with either the protanope or deuteranope (Fig 8).

The visibility of normal trichromats to fluorescent colors was consistent with the reported field-based visibility, with orange being the most visible fluorescent color. However, no difference was observed between fluorescent colors, but was comparable to fluorescent yellow [4,11]. These color visibilities are roughly in line with the specific spectral sensitivity reported by the Commission Internationale de l’Eclairage, rather than the luminance of the cloth [12]. Fluorescent orange, which is the most visible color to people with normal trichromats, showed similar results in deuteranopes. However, fluorescent yellow was similar in protanopes, but fluorescent orange was less visible, with the same average degree of preference as the less brightly perceived white/short-wavelength and black combined samples. Protanopes lack long wave-sensitive cones [13], which suggests that long-wavelength colors such as red and orange are perceived as darker, and not only fluorescent orange but also the visibility of the combination of white and red had as lower average degree of preference than normal trichromats. Therefore, protanope visibility should be a primary consideration when producing color designs; yellow is most suitable, and fluorescent yellow is even better.

The visibility of conspicuousness sensitivity in color vision defect is very similar to the results we previously reported for normal trichromats [2]. In people with a color vision defect, the larger the contrast difference in brightness between the two-color combinations, the higher the visibility result.

Considering both protanopes and deuteranopes, black/fluorescent yellow was found to be the most visible, followed by black/yellow, black/white, black/yellow-red, and white/red-purple; all were equally visible. All of them combine high-and low-brightness colors. Short-wavelength colors were found to be incompatible with black. Combining the two colors with a contrast difference proved to be the best way to improve visibility.

Our study was conducted in a sunset environment. Therefore, the recommended clothing colors are not intended to be effective at night when dark. Rather, our aim is for optimizing the clothing design of school-age children for the situations of traveling to and from school in situations where there is natural light, such as at sunset. In this study, we found that black and fluorescent yellow had the highest visibility, with a contrasting difference between the two colors. It is suggested that wearing clothing of these colors during the day and at sunset improves visibility. Retroreflective materials are not the most visible during the day because they are less luminous than fluorescent yellow or yellow fabrics. At night, retroreflective materials are maximized by automobile headlights; therefore, the use of conventional high-visibility safety clothing is recommended. It is anticipated that the safest approach would be to use clothing based on the results of this study during the day and at sunset, and wearing conventional clothing at night.

A person wearing this combination of clothing during the day or at sunset is easily recognized as an object; however, to be recognized as a person, indirect aspects such as "biological movement" need to be emphasized, and this should be considered when creating high-visibility safety clothing for testing [14].

A limitation of this study is that most of the participants in the present study were middle-aged in their 40s–50s with congenital color vision defects, and the cone contrast test showed a slightly reduced S-cone sensitivity compared to younger participants (S2 Fig). However, since none of the participants had serious acquired or congenital color vision defects in the sensitivity of S-cones, and as per previous reports, the total transmission of light was minimally affected in the 40–59 age group [15]. Further, we believe that the effect of age was small in the present study.