Exploration of visual variable guidance in outdoor augmented reality geovisualization

ABSTRACT

The visual perception of augmented reality (AR) geovisualization is significantly different from traditional controllable 2D and 3D visualization. In this study, we extended the rendering styles of color variables to include natural material color (NMC) and illuminating material color (IMC) and extended the size to include linear size (LS) and angular size (AS). Outdoor AR geovisualization user experiments were conducted, examining the guidance characteristics of five static variables (NMC, IMC, shape, AS, LS) and two dynamic variables (vibration, flicker). The results showed that the two dynamic variables provided the highest guidance, and among all the static variables, the order of guidance was shape, IMC, AS, NMC, and finally LS. This is a new finding that is different from the color, size, and shape guidance order in 2D visualization and the color, shape, and size order in 3D visualization. The results could be utilized to guide the selection of visual variables for symbol design in AR geovisualization.

KEYWORDS:

• Visual guidance
• visual variable
• augmented reality visualization
• virtual geographic environment

1. Introduction

Geovisualization is an intuitive and practical approach to communicating abstract geographic information using graphical symbols. As an interface between user and spatial information, a high-availability geovisualization system should help users efficiently and accurately to identify target symbolics from a visualization context (Swienty et al. 2007). According to Bertin’s visual variable theory, graphical symbols can be decomposed into seven essential visual variables: position, size, shape, color (hue/value), orientation, and texture (Bertin 1983). This is congruent with the human brain's visual perception and cognitive mechanism, in which visual features such as color, size, or orientation are perceived and processed in different visual cortex areas (Swienty et al. 2007). Therefore, differences in visual perception mechanisms will result in different perceptual efficiencies among these seven visual variables (Mao 2008; Neisser 2014; Yang and Li 2021). The use of visual variables to guide one’s visual attention to facilitate searching for a target object is known as visual guidance (Liu, Dong, and Meng 2017). Considering the difference in the perception mechanism of visual variables and different rendering effects of visualization media, the guidance provided by a given visual variable may vary across geovisualization conditions (Stachon et al. 2018).

The geovisualization form and medium can affect the expressiveness of visual variables (Carpendale and Sheelagh 2003), which is an essential factor in determining a visual variable’s guidance. Williams analyzed eye fixations and response time in visual search tasks. Target symbols of different color, size, and shape visual variables were scattered in a 2D plane. The results showed that color provides the highest level of guidance, followed by size, with shape providing the least. (Williams 1966). Garlandini conducted a controlled experiment to investigate the perceptual salience of four visual variables in 2D static maps under a flicker condition, and the results showed that the size and color value could guide a participant’s visual attention better than color hue and orientation (Garlandini and Fabrikant 2009). Wolfe's summary survey of visual variable guidance led to similar conclusions: color and size provide high levels of guidance, while shape is weaker in the context of 2D visualization (Wolfe and Horowitz 2004).

In a 3D geovisualization context, Yang assessed the visual attractive performance of color, size, and shape with 3D projected images captured from a street view and a 3D map. The results indicated that the guidance order is color, size, and shape, which is the same as Garlandini's result obtained in a 2D static map experiment (Yang and Li 2021). This is probably because the visual perception of the experimental material is more likely a 2D image than a 3D scene. Similar results were obtained from other 3D and virtual reality visualization studies: color exhibits a higher and more stable guidance property (Arjun et al. 2021; Pouliot et al. 2014). Although Liu conducted experiments in scenes where the 3D perspective was more pronounced, he evaluated the guidance and constancy properties of size, shape, and color in 3D visualization. The stimulus materials were screenshots of a true-3D scene under default lighting settings taken with a virtual cameral with a 35° field of view (FOV) located at a 50-meter height position, and the results showed that color and shape provide high-level guidance. Size provides limited guidance; however, it provides predominant guidance in 2D visualization. In addition, size constancy was perceived with lower accuracy than color and shape (Liu, Dong, and Meng 2017). This indicates that the introduced depth cues in the perspective view in 3D visualization can affect the perception of some visual variables.

With the development and maturity of augmented reality (AR) technology, AR devices are commercially available in various domains. As a new visualization medium, AR provides a promising approach to geovisualization by creating a mixed visualization environment that superposes a virtual visualization scene onto its representation environment (Dickmann et al. 2021; G. Zhang et al. 2020). Compared with traditional 2D/3D geovisualization, AR geovisualization focuses on transferring geographic information to users in real time, which requires visualization symbols constituted by visual variables to be quickly perceived (Reichenbacher 2004). MacEachren also mentioned that in the immersive visualization environment of wearable devices such as AR and virtual reality, expressing and interacting with geographic data efficiently are problems that need to be further investigated (MacEachren 2019).

The virtual scene and natural environment are perceived synchronously within a limited FOV in AR geovisualization, which can increase the received information amount and affect the visual perception efficiency of the mixed geovisualization environment. Gabbard conducted visual search experiments in AR visualization to investigate search efficiency in various backgrounds, such as brick, buildings, sidewalks, and sky, and the results showed that the background type significantly affects the search efficiency (Gabbard et al. 2007). This suggests that visual attention in AR visualization is influenced by both virtual content and the background environment (Dong et al. 2021; Duan et al. 2022). In addition to environmental factors, the narrow FOV of AR devices is a crucial factor that significantly affects search performance, especially for individuals wearing head-mounted display (HMD) AR glasses (Trepkowski et al. 2019). In addition, color distortion caused by monitor and natural light interference may lead to inaccurate color perception, which also results in reducing the perception efficiency of color in AR (Livingston, Barrow, and Sibley 2009). Halik systematically studied the level of usefulness of static visual variables in mobile AR cartographic design. He concluded that as the observing perspective changes from a geocentric view to an egocentric view, the approach to the presentation of point symbols also needs to be changed adaptively (Halik 2012).

In summary, previous studies have focused more on traditional 2D and pseudo-3D visualization, which are quite different paradigms from AR geovisualization. The obtained conclusions are not applicable to AR and virtual reality geovisualization (Herman et al. 2018; Meng 2005). Although some studies have investigated the effect of various factors, such as the environment, FOV, and rendering distortion, on AR perception efficiency, the effectiveness and guidance provided by visual variables have not yet been systematically studied, especially in an outdoor AR environment. Therefore, in this study, we design and administer user-participation experiments utilizing HMD AR devices in an on-site outdoor environment to evaluate the guidance performance of visual variables in AR geovisualization.

2. Visual variables in AR

2.1. Characteristics of AR geovisualization

AR is a visualization and interactive system that creates a virtual-real mixed 3D environment by registering digital virtual objects into a real environment (Azuma 1997). AR visualization integrates visualization scenes and the real environment and builds a continuous spatial cognition and interaction environment, avoiding interrupting the flow of geographic information perception when interacting with the mixed environment and improving information transmission efficiency (Carbonell Carrera and Bermejo Asensio 2017; Ma, Zhang, and Huang 2021). Therefore, AR geovisualization has been widely adopted in fields such as battlefields, emergencies, and walking navigation (Werner 2019; You et al. 2018).

Handheld and HMD AR devices are currently the two primary commercial AR technology categories. Handheld AR devices, such as smartphones and tablets, render mixed environments through a video stream on a 2D screen, making the render effect pseudo-3D and closer to a typical mobile map. HMD AR renders mixed environments by blending the light from the outside environment and the AR screen, making it more consistent with how people naturally perceive their surroundings. The overlap of visualization and interaction spaces in HMD AR geovisualization prevents interruption of spatial perception and cognitive processes during interaction. This results in some distinctive AR geovisualization characteristics. Some of these characteristics are derived from the abovementioned AR perception and cognition features, and others are derived from the performance limitations of existing AR devices.

(1) User-centered immersive visualization view

In traditional 2D and 3D maps, users are located outside the visualization space and observe the map and geovisualization from a third-person perspective. In essence, both 2D maps and screen-based 3D maps are perceived as interactable 2D images. In AR geovisualization, the user can walk into the visualization scene and become part of it. Therefore, the visualization view correspondingly changes from geocentric to egocentric (Meng 2005). In AR geovisualization, all symbols are arranged around the user, and the mixed visual environment is perceived from an immersive binocular stereo first-person perspective, which makes depth cues an essential factor affecting the expression of visual variables, such as symbols with the same size being perceived as having different visual sizes at different viewing distances.

(2) Dynamic visualization background

Similar to a ‘basemap’ in traditional maps, the real world serves as an information source in AR geovisualization, providing spatial reference and information supplementation for the virtual geovisualization scene. However, for a mixed scene, the real environment is uncontrollable. Therefore, the real environment not only enhances the virtual scene but also may obstruct and restrict the expression of scene rendering. For example, dynamic environment lighting can affect the perceptibility of virtual scenes (Kahl, Ruble, and Krüger 2022). Outdoors, where the natural background light is strong, the visibility of the virtual scene may be reduced, resulting in In outdoor environments with strong natural light, the visibility of the virtual scene may decrease, resulting in a loss of visual detail for symbol perception. The variety of colors in an environment will also affect the consistency of color perception in AR (Gabbard et al. 2022). The perceived color is not determined by only the color value in AR but is the combined effect of color value, transparency, shadow, light, and the natural environment background. In addition, dynamic objects in the environment such as pedestrians, vehicles, and flashing lights are more likely to attract the user's attention than the static symbols in the mixed scene (Rautenbach et al. 2015), which could result in distraction from the mixed scene to irrelevant dynamic objects.

(3) Narrow field of view

With the inherent performance constraints of the existing AR devices, the AR FOV is still much narrower than the human eye FOV. The advanced HoloLens 2 AR device has a diagonal FOV of approximately 52°, which can cover only the near peripheral perception area of ⁣⁣the eyes (McKeefry, Murray, and Parry 2007). During use, the mixed scene is scanned by sliding windows rather than an overall view of the full extent of digital maps displayed on a computer or smartphone screen. This requires that a target object be efficiently perceived during the movement and rotation of the virtual scene driven by user interaction, especially in a target identification task, which is the basic operation in AR geovisualization (Kruijff et al. 2019). This issue is caused by technical limitations. With the advancement of AR technology, the narrow FOV problem will eventually be solved, but it is indeed a nonnegligible limiting factor restricting AR geovisualization at present.

2.2. Extension of visual variable rendering style

Bertin's visual variable theory was proposed in the context of paper maps (Carpendale and Sheelagh 2003). With the advancement of cartographic technology over the decades, the visual expressiveness capability of maps has also been improved, and a variety of static and dynamic visual variables have been proposed. (MacEachren 2019; 2004; Slocum et al. 2022). However, those complemented variables are mainly suitable for 2D or screen 3D digital maps, not AR geovisualization. Therefore, we extend the rendering style of color and size visual variables to meet the characteristics of AR geovisualization.

2.2.1. Color rendering extension

In conventional maps, the color value is represented by RGB on a screen and CMKY on a printed map. Therefore, once a color value is set, the color rendering effect is also determined, which will only be slightly affected by the color difference of the screen or printer. However, in AR geovisualization, the color acts only as an original value of the rendering effect, and the final rendering effect needs to be calculated in real time based on the light condition, materials, environmental features, and camera position in the scene. If shadows and light reflections are considered, it will be more complicated. This causes the same color value to be rendered differently in an AR geovisualization scene under different rendering conditions.

Among these rendering factors, the material is the dominant factor affecting the color rendering effect. We divide color variables into two categories based on how the light affects rendering: natural material color (NMC) and illuminating material color (IMC). The former calculates the color by reflecting the light source in the scene, similar to the color rendering on a paper map; the latter uses the color value as a light emission color parameter to render the color, similar to the color rendering mode of a digital map. Figure 1 shows the rendering result of two yellow (color value: #FFFF00) materials.

Figure 1. Natural and illuminating material color, (a) NMC, (b) IMC.

2.2.2. Size rendering extension

The size visual variable is measured with pixels or millimeters in a 2D map, which has strong stability and will not change with the map reader viewing position. In 3D AR geovisualization, the scene is viewed in an egocentric perspective, and the visual size of the symbol is determined by not only the size of the symbol but also the distance from the observer. Size variables measured in pixels or meters would lose their ordered perception, which makes them unsuitable for AR geovisualization in some situations, such as expressing gradient information. Therefore, we propose an extended form angular size (AS) for size variables that measures size in angles, which is different from the traditional linear size (LS) in meters and pixels. As shown in Figure 2, the size of the object in the figure is ${\alpha }^{\circ }$, and the diameter of the symbol increases with the distance from the observer, but the perceived visual size of the three symbols remains the same.

Figure 2. Schematic of angular measurement size.

2.2.3. Motion enhancement

In geovisualization with 2D and screen 3D maps, the background of cartographic symbols is static and controllable, and the attention allocation of visual perception during map reading is mainly focused on the symbols constituted by various visual variables. In AR geovisualization, the visualization scene is mixed with the real environment, making the visualization environment more complicated. Bright light can obscure the virtual scene, reducing the salience of visual variables. The real environment changes dynamically, and various dynamic elements, such as pedestrians and vehicles, may distract one's attention and affect the perception of static visual variables. Therefore, to enhance visual guidance, static visual variables can be supplemented with dynamic effects to create dynamic visual variables (Pouliot et al. 2014). However, it should be noted that since dynamic visual variables themselves can also express semantic information, such as the changing speed or a phenomenon's developmental process, the introduction of dynamic variables may reduce the effectiveness and efficiency of information expression (Bertin 1983; Köbben and Yaman 1995). However, because the AR geovisualization context is a highly dynamic environment, the high-level attention-attracting characteristics of dynamic variables can help users efficiently identify symbols in complex real environments (Koussoulakou and Kraak 1992). Therefore, vibration and flicker dynamic elements were selected to improve the visual guidance provided by static visual variables in this study.

In summary, in AR geovisualization, the visualization environment and the perspective of observation are significantly different from those of traditional maps or geovisualization, which may result in visual variables having different guidance features. As a result, user experiments are required to investigate the guidance provided by visual variables in AR geovisualization.

3. Method

In this study, we utilized user experimentation to investigate guidance of visual variables in outdoor AR visualization. The overall technical roadmap for this study consists of two parts: experimental design, represented by the green block in Figure 3, and data analysis, represented by the red block in Figure 3. The experiment includes three stages: preparation, practice, and execution. The flowchart provides a brief description of the experimental task settings, sequence, and scenes used. The data analysis section describes the indicators and methods used for analyzing the eye-tracking and task data obtained during the experiment. Specific details such as the design of the experimental scene, procedures, and data analysis methods will be discussed in subsequent sections.

Figure 3. Experimental and data analysis framework.

3.1. Experimental design

The primary objective of this research is to investigate the visual guidance properties of commonly used visual variables and their extended rendering style in an outdoor AR visualization environment. These visual variables include seven rendering forms, the five static visual variables (IMC, NMC, shape, AS and LS), and two dynamically enhanced forms (vibration and flicker). Identification is a simple and elementary task in geovisualization that has been employed to evaluate visual guidance in 2D and 3D geovisualization (Harada and Ohyama 2022; Liu, Dong, and Meng 2017). Therefore, identification is selected as the experimental task, including seven subtasks in seven scenes corresponding to the abovementioned visual variables. The task details are shown in Table 1.

Table 1. Tasks in the visual guidance experiment.

Task	Visual variable	Details
Task 1	NMC	Find and select all the red symbols rendered in NMC from Scene 1
Task 2	IMC	Find and select all the red symbols rendered in IMC from Scene 2
Task 3	Shape	Find and select all the spherical symbols from Scene 3
Task 4	LS	Find and select all the symbols larger than the other normal-size symbols from Scene 4
Task 5	AS	Find and select all the symbols larger than the other normal-size symbols from Scene 5
Task 6	Flicker	Find and select all flickering symbols from Scene 6
Task 7	Viber	Find and select all vibrating symbols from Scene 7

Experiments were conducted in an outdoor campus setting, which included landscapes such as buildings, roads, and trees and dynamic objects such as pedestrians and vehicles. Since the lighting conditions could interfere with a participant’s perception of visual variables, to guarantee that the lighting conditions of all participants were consistent, we conducted the experiment between 15:30 and 16:00, when the lighting conditions were softer than at midday. Each task had a maximum completion duration of 120 s. The experiment automatically moved to the next experimental task when one task was completed or timed out. On average, each participant required approximately seven minutes to complete all tasks.

3.2. Experiment scene

3.2.1. Visual variable setting

In our experiments, visual variables in seven rendering forms were utilized in the identification tasks. As shown in Table 2, the color variable includes five colors that were rendered as NMC and IMC materials. The shape variable included four simple 3D geometries: sphere, cube, cone, and capsule. The size variable included two forms: AS and LS. Vibration was achieved by periodically changing the object position relative to the scene's origin. Flicker was achieved by changing colors periodically.

Table 2. Visual variable values.

Variable	Properties	Examples
NMC	White: #FFFFFF Green: #00FF00 Blue: #0000FF Yellow: #FFFF00 Red: #FF0000 Light Color:# #FFF4D6 Light Intensity:1 Shadow Type: Soft Shadow
IMC	White: #FFFFFF Green: #00FF00 Blue: #0000FF Yellow: #FFFF00 Red: #FF0000
Shape	Sphere Cube Cone Capsule
LS	Size measured in length
AS	Size measured via angle
Flicker	Green and white flash alternately with a period of 0.2 s
Vibration	Vibrate at a speed of 0.1 m/s within 0.05 m in the vertical direction (along the Y axis).

3.2.2. Scene design

In accordance with the experimental design, seven experimental scenarios, Scene 1 through Scene 7, were created. As illustrated in Figures 4a and b, all seven scenes are based on a basic sketch scene, in which all symbol positions are preset within a range of 50 m horizontally wide and 30 m vertically high. (Figure 4b). The experimental scene consists of 26 symbols, six of which are targets to be identified (such as the red object in Figure 4a), while the remaining twenty symbols serve as backgrounds (such as the white object in Figure 4a). To maintain an even distribution of symbols, all symbols are scattered throughout 180° in front of the participants, and each 30° area contains only one target. The placements and quantities of symbols in Scenes 1–7 are consistent with the basic sketch scene. To avoid a scene-learning effect, the type and color of symbols at the same position in different experimental scenes are varied, and these variations were selected from the values listed in Table 2. For example, in Scene 1, a position contains a blue cube, which is replaced by a yellow cone in Scene 2.

Figure 4. Guidance experiment scene: (a) sketch of the horizontal distribution of symbols, (b) sketch of the vertical distribution of symbols, (c) shape and IMC variables adopted in Scenes 4 and 5, (d) an example of symbol effects employed in other scenes.

All color and shape variables in Table 2 are adopted in all scenes except for Scene 4 and Scene 5 (Figure 4d), and the size of all symbols are measured via angle. The colors in Scene 1 are rendered with NMC materials, and the other scenes are rendered with IMC materials. Since color and shape can affect size judgments, only white spheres are used as experimental symbols in Scenes 4 and 5 (Figure 4c), since color differences would affect size judgments of symbols (X. Zhang et al. 2018), and symbols of different shapes could not be compared in relative size. In Scene 4, the target spheres are 1.2°, and the others are 1°. In Scene 5, the target spheres are 1 m, and the others are 1.2 m. Scenes 6 and 7 are similar to Scene 2 but include flicker or vibration to the target symbols.

3.3. Apparatus

The experiment utilized a Microsoft HoloLens 2.0 as the AR device. The device has a diagonal field of view of 52° with 2k 3:2 resolution and supports eye-tracking technology, which can record user eye movement data at a rate of 90 Hz using the Unity 3D Extended Eye Tracking API. The device supports voice input and output, which could be used to play voice narrations of experiment guidance information using text-to-speech (TTS) technology.

3.4. Procedures

Thirty-two participants (15 females and 17 males) were invited to participate in the experiment. They were postgraduate students majoring in geographic information system (GIS) or remote sensing (RS), with an average age of 24.9 (SD = 1.88). No participants had experience using head-mounted AR. All participants had normal or corrected-to-normal vision. After completing the experiment, each participant received 70 yuan as an experimental reward.

We used two HoloLens devices and invited four participants to conduct experiments each day. The experiment took a total of nine days, including one day for testing equipment and scenarios and eight days for formal experiments.

The experiment is divided into three stages: preparation, practice, and execution. During the preparation stage, the basic operation of the experiment's equipment is explained, and visual calibration is performed for each participant using an in-built calibration application. In the practice phase, the participants learn to use the cursor to gaze on and select target symbols; they are instructed not to walk but to just turn their heads to identify objects. During the execution phase, the experiment is automatically directed by the experiment system based on the task flow, with audio narrations describing each task to the participants. For instance, before Task 1 begins, the system announces, ‘Please select all red symbols from the scene.’ Eye movement data, subject operations, and task completion results are recorded in real time during the experiment.

To prevent participant movements from influencing the trial outcomes, only head rotation is permitted, and free walking is prohibited. In terms of the order of the tasks, first, Task 1, Task 2 and Task 3 are presented in a random order, followed consecutively by Task 4 and Task 5, and finally Task 6 and Task 7 in a random order. The random task order is to avoid a learning effect caused by the participants becoming familiar with the experimental scene based on a fixed order. The symbol is selected by placing a head-gaze cursor on it and then confirming the selection after 1.5 s.

3.5. Evaluation indices

In the experiment, task finish time (FT), time to first fixation (Ttff), visit ratio (VR), and accuracy (AC) are adopted as the indices of guidance evaluation, and these indicators have been proven to be effective for visual variable guidance evaluation in 3D geovisualization scenes (Liu, Dong, and Meng 2017).

The FT is the total time spent completing each task, and the guidance is negatively correlated with the FT; that is, the shorter the FT is, the greater the visual variable can attract the participant's attention during the visual perception process. The Ttff is similar to the FT, and the shorter the fixation time is, the higher the level of visual guidance. The VR is calculated as the ratio of the gaze duration fixed on the identification symbols to the total gaze duration. A higher VR indicates that the participant devoted more attention to the target symbol than to other interfering background objects. The AC evaluates guidance from the perspective of task completion quality, and a higher accuracy indicates a higher level of visual guidance.

To ensure the reliability of the results, we conducted a guiding evaluation using multiple indicators. However, due to the inability to directly calculate different evaluation indicators, we first calculated the guiding order value for each variable in each indicator. The calculation of the order value involved sorting the visual variables according to their guidance in the indicator, where higher guidance resulted in a smaller guiding order. The first visual variable was assigned a ranking of one. Subsequently, we compared each adjacent visual variable. If the indicator values of the next variable significantly differed from the previous variable, we increased the order value by one. If there was no significant difference, we assigned the same order value as the previous variable. For instance, for the FT indicator, both vibration and flicker had the highest guidance with no significant difference between each other, resulting in a shared order value of one. In contrast, the shape indicator value was significantly higher than that of vibration and flicker, resulting in an order value of two. After setting the order values for all visual variables, we obtained a comprehensive evaluation of the guidance of each visual variable in all indicators. Finally, the guidance order ($G_v$) of each visual variable was calculated based on Formula (1), with smaller $G_v$ values indicating higher guidance. (1) $G_v=O_v^{FT}+O_v^{VR}+O_v^{Ttff}+O_v^{AC}$(1) In Formula (1), $G_v$ is the overall guidance of a given visual variable. A lower $G_v$ indicates a higher guidance associated with the variable. $O_v^{FT}$, $O_v^{VR}$, $O_v^{Ttff}$, and $O_v^{AC}$ correspond to the order value of the visual variable in relation to the four evaluation indices, FT, VR, Ttff, and AC, respectively.

4. Results

In this section, we report the visual guidance performance of the seven visual variables. In general, the guidance provided by color, size, and shape visual variables showed some differences from traditional 2D and screen 3D visualization, and the detailed results are as follows.

4.1. FT performance

Figure 5a shows the time consumption of the seven visual variables, and Figure 5b shows the Mann‒Whitney significance test results of each pair of visual variables. The results show that the shape visual variable FT was the shortest and significantly differed from NMC (FT_shape-NMC= −4.98, p = 0.044), IMC (FT_shape-IMC= −2.6, p = 0.041), and LS (FT_shape-LS= −81.78, p = 0). The AS did not show significant differences from either the NMC or IMC variables. The IMC FT was 2.38 s less than that of NMC, but the difference was not significant. The LS FT was up to 111.7 s and was significantly longer than that of the AS (32.1 s). In terms of the dynamic visual variables, the results showed that the FT was significantly shorter than that of the five static visual variables. However, there was no significant difference between vibration and flicker, with both having an FT of approximately 22.5 s.

Figure 5. (a) Statistics and (b) significance testing results of the FT.

4.2. Ttff performance

The Ttff and Student’s t test statistical results are shown in Figure 6a and b. Among all the static visual variables, the IMC Ttff was 3.38 s, which was the shortest and significantly shorter than the NMC Ttff (Ttff_IMC-NMC = −1.56, p = 0.018) and AS Ttff (Ttff_IMC-AS= −2.44, p = 0.07) variables. The shape variable was slightly higher than the IMC variable (Ttff_shape-IMC= 0.66, p = 0.799), but the difference was not significant. The AS took approximately 5.82 s, which was significantly higher than the IMC and shape variables, and the LS took the longest time at approximately 33. 86 s. In terms of the dynamic variables, both vibration and flicker generally had a shorter Ttff than static variables. However, it should be noted that the IMC Ttff took slightly less time than the flicker variable. Although the difference was not significant, it suggests that flicker had no dominant advantage in visual guidance over IMC. The vibration Ttff took significantly less time than that of flicker, but the advantage over IMC was not significant (Ttff_{vibration-IMC}= −0.32, p = 0.072).

Figure 6. (a) Statistics and (b) significance testing results of the Ttff.

4.3. VR performance

The VR and Student’s t test statistical results are shown in Figure 7a and b. Among all the static variables, the IMC, shape, and AS VRs were roughly equivalent at approximately 0.3, with insignificant differences. All VRs of the above three variables were significantly higher than those of NMC and LS. In terms of the dynamic variables, the vibration VR was the highest and significantly differed from the other variables. The flicker variable also showed a higher VR, but it was not significantly different from either the IMC or shape static variables.

Figure 7. (a) Statistics and (b) significance testing results of the VR.

The eye saccade point (ESP) distribution (green points) also yielded a similar result to the VR index, as shown in Figure 8. Compared with the static visual variables, the ESPs of the dynamic variables were more concentrated near the identification symbols (red points). In terms of static variables, compared with IMC, more ESPs are scattered in nonidentification symbols (blue point) regions in NMC, indicating that IMC provided superior guidance to NMC. Compared with the AS, a large number of ESPs in the LS were distributed around nonidentification symbol regions, indicating that the guidance provided by the AS was better than that of the LS. The distribution of saccade points was generally similar among IMC, shape, and AS, which is consistent with the results in the VR metric.

Figure 8. Eye saccade point distribution map of visual variables.

4.4. Accuracy of statistical results

The AC and Mann‒Whitney significance test results are shown in Figure 9. The average AC of the visual variables was higher than 0.9, except for the AS and LS size variables. The shape (AC = 0.99) had the highest accuracy, which was significantly higher than other variables except for flicker. Although there was no significant difference between NMC and IMC, the box plot shows that the IMC accuracy data distribution tended to be more concentrated and stable. The accuracy of the AS was significantly lower than that of other variables except for the LS, which was only approximately 0.87. The accuracy of the LS (AC = 0.22) was even lower than the probability of random selection of 6 from all 26 symbols (0.23).

Figure 9. (a) Statistics and (b) significance testing results of the AC.

To express the guidance characteristics of all visual variables concretely, we calculated the guidance order of visual variables on the FT, Ttff, VR and AC evaluation indicators and then calculated $G_v$.

As shown in Table 3, vibration has the lowest $G_v$ value, which means it has the highest guidance, while LS has the lowest guidance because of its lowest $G_v$ value.

Table 3. Guidance orders and values of visual variables.

	Vibration	Flicker	Shape	IMC	AS	NMC	LS
$O_v^{FT}$	1	1	2	3	3	3	4
$O_v^{VR}$	1	2	2	2	2	3	4
$O_v^{Ttff}$	1	2	2	2	3	3	4
$O_v^{AC}$	1	1	1	2	2	2	3
$G_v$	4	6	7	9	10	11	15

5. Discussion

5.1. Guidance provided by dynamic visual variables

Adding vibration and flicker visual variables to static visual variables can enhance guidance by introducing dynamic attributes. Considering the four experimental indices comprehensively, the guidance provided by dynamic visual variables was superior to that of static variables. Specifically, the vibration variables were significantly better than all static variables in terms of the FT and VR indices, although the difference in the Ttff was not very significant (p = 0.07). The boxplot figures showed that the data distribution of all vibration indices was more concentrated than the other visual variables, indicating that the guidance performance of this variable was more stable across all subjects. Although the flicker variable was generally superior to the static visual variables, it did not demonstrate significant advantages over the IMC and shape variables in the VR and Ttff indices. This is because the flicker is composed of two periodically changing IMC variables; thus, it is similar to IMC in visual perception.

Comparing the two dynamic variables, the vibration was significantly better than the flicker on the VR and Ttff indices. The vibration variable had the highest VR, indicating that it effectively captured a participant’s attention, particularly for symbols on the periphery of the field of view, and that vibrating symbols can be detected rapidly (Hansen, Pracejus, and Gegenfurtner 2009; Matsuzoe et al. 2017), which allowed a subject to quickly identify the target outside the macular area. Therefore, the vibration variable had an advantage over the flicker variable.

Generally, in outdoor AR geovisualization, the high level of guidance provided by dynamic symbols also requires users to allocate a significant amount of attention. This affects user perception of the external environment and threatens their safety. Therefore, it was inappropriate to place numerous dynamic symbols in an AR geovisualization scene, which is better suited for the prominent expression of a single or a small number of important points of interest (POI) symbols.

5.2. Guidance provided by static variables

Among all the static visual variables, both IMC and shape provide stronger guidance compared to the others, and the shape performed better than IMC on the FT indices because it is less time consuming, which is different from the conclusion in a previous 3D visualization guidance experiment that color is less time-consuming than shape (Liu, Dong, and Meng 2017). The primary reason is that the color in AR geovisualization is readily affected by the interference of background and lighting conditions in the real outdoor environment, as well as the color rendering distortion of the AR equipment, which reduces the judgment of a subject on the color variable. The same trend is also implied by the AC index for shape being significantly higher than that for IMC. Therefore, the guidance provided by IMC may not be as dominant as in traditional 2D and 3D visualizations that are rendered on a static background.

The guidance provided by IMC is stronger than that provided by NMC in terms of performance on all indices. Since IMC is not affected by the lighting and shadows of the virtual scene, the IMC-rendered color saturation and brightness are higher than NMC with the same color value, which results in an IMC-rendered symbol being easier to correctly perceive, and the AC index also reflects a similar trend. We surveyed the participants and found that when the symbol was at the edge of the visual field, NMC color rendering was easily affected by the shadow produced by the main light source and the color distortion caused by the AR display, leading to a situation where the yellow symbol appeared closer to red and could be easily confused or misinterpreted. NMC can be rendered more realistically by keeping the lighting conditions similar to those of a real environment, which could also reduce the level of visual guidance as it is easily affected by the intensity of external light. Therefore, the NMC variable is more suitable for rendering realistic models that aim to achieve a high level of immersion by blending virtual symbols with the real environment, thus potentially enhancing the overall user experience and the sense of presence within the AR geovisualization. For example, the use of NMC color on building models is an effective way to showcase their real-world appearance, as they respond to different lighting conditions throughout the day. In regard to rendering abstract symbols, IMC is the better choice due to its high stability of rendering effect under dynamic lighting conditions. Its brightness allows it to resist the impact of the primary light source, resulting in a more consistent rendering effect compared to NMC.

Among the two size variables, the AS also showed moderate guidance, outperforming NMC on the VR indices, but its accuracy (0.87) was lower than that of the other variables, except for the LS. Due to the lack of depth cue assistance, the size of the symbol represented by the LS was unable to be accurately perceived by the participants, and all evaluation indices indicate that the LS variable provided almost no visual guidance. However, this does not mean that LS is useless in AR geovisualization. The visually perceived size of LS symbols is consistent with human perception of real objects distributed in the environment. Therefore, LS is similar to NMC and is suitable for realistic symbol rendering, while AS is more suitable for the expression of abstract symbols that require high perceptual consistency. This consistency relates to the need for symbols of the same type and meaning to have the same perceptual size. For instance, in the case of using a triangular cone symbol to indicate the location of all nearby hotels and where there is no need to emphasize a particular hotel, the perceptual size of these symbols should remain uniform.

5.3. Limitations

The empirical results reported herein should be considered in light of some limitations. The experiment only used simple abstract symbols and did not consider complex and textured symbols, which may have different guidance effects compared to simple symbols. Some studies have suggested that model complexity can be considered a visual variable (Rautenbach et al. 2015). Additionally, to ensure an even distribution of the symbols, they were not spatially associated with the real environment, which may have affected the perception of the symbols. Moreover, the experiment did not test the guidance in different complex background environments, as the complexity of the background can also affect the efficiency of symbol perception (Stachon et al. 2018). Last, due to the limitations of existing AR display technology, the realism and immersion of the virtual-real mixed scene were low, which could also affect the guidance of visual variables.

Hence, to address the limitations of this study, future research should focus on investigating the guidance effects of visual variables under more complex environments and symbols. By comparing and analyzing the effects of background and symbol complexity on the guidance effects of visual variables in contrast to the conclusions of this study, a more comprehensive understanding of the relationship between visual variables and perception can be gained.

6. Conclusion

In this study, we analyzed the characteristics of AR geovisualization compared with traditional 2D and 3D maps and visualization and extended and enhanced the rendering styles of visual variables in AR geovisualization. Color was extended to IMC and NMC, size was extended to LS and AS, and vibration and flicker dynamic enhancements were performed on static visual variables. A user experiment was carried out in an outdoor environment to evaluate the visual guidance provided by these variables.

Based on the $G_v$s values from Table 3, the guidance order of the seven testing visual variables is as follows: vibration, flicker, shape, IMC, AS, NMC and LS. This is a new finding that is different from the color, size, and shape guidance order in 2D visualization and the color, shape, and size guidance order in 3D visualization. The results can be utilized to guide the selection of visual variables for symbol design in AR geovisualization. The dynamic visual variables of vibration and flicker provide the highest guidance, but vibration is more stable and effective than flicker if only one symbol can be perceived in the FOV at a time. Although flicker overall provides higher guidance than shape, there is no significant difference in other indicators apart from the FT index. Among the static visual variables, shape surpasses IMC in guidance, providing the highest guidance. This is different from 2D and 3D visualization, where color can provide the highest guidance (Wolfe and Horowitz 2004) because color perception in AR visualization is readily disturbed by the external environment (Livingston et al. 2013). Size can also provide high guidance in 2D visualization (Garlandini and Fabrikant 2009), but its guidance is limited in 3D (Liu, Dong, and Meng 2017). However, our research has found that traditional size, or LS in this paper, can no longer provide effective guidance in AR geovisualization, while AS still provides guidance, and its guidance is higher than that of NMC.

Based on the guidance characteristics of the visual variables mentioned above, the following are some suggestions and guidelines for symbol design in AR geovisualization: (1) Size is a necessary attribute of symbols, but since LS no longer provides guidance in HMD AR geovisualization, AS should be preferred if the absolute size of the symbol is not necessary. (2) Color is also an important material for symbol design. In AR geovisualization, with the same color, the saturation and brightness rendered by IMC are higher than those rendered by NMC. This is an important advantage in outdoor dynamic lighting and background environments. Therefore, IMC should be prioritized unless it is necessary to express the mixed effect of the symbol with the real environment. (3) Dynamic visual variables can provide the highest guidance, which is suitable for indicating important targets, such as highlighting targets that meet search criteria from multiple symbols. Since vibration is more advantageous than flicker in terms of guidance, it is recommended to use vibration to enhance the expression of symbols. (4) The NMC and LS variables were preferred when expressing realistic symbols and models.

Geolocation information

The study area of this research is an outdoor environment and the location where the experiment was conducted is located in Chaoyang District, Beijing, China.

Acknowledgments

The authors would like to thank all participants in the experiment and the editor and anonymous reviewers for their careful reading and constructive suggestions.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data that support the findings of this study can be obtained from the corresponding author upon reasonable request.

Additional information

Funding

This research was supported and funded by the Pilot Fund of Frontier Science and Disruptive Technology of Aerospace Information Research Institute, Chinese Academy of Sciences [Grant number E0Z211010F].

Appendices

Table A1. Abbreviation list.

Abbreviation	Full name	Abbreviation	Full name
NMC	Natural material color	Ttff	Time to first fixation
IMC	Illuminating material color	VR	Visit ratio
AS	Angular size	AC	Accuracy
LS	Linear size	$G_v$	Guidance order
FT	Task finish time	POI	Point of interest