2. Literature
With the arrival of the smartphone variant of the mobile phone in 2007, pedestrian wayfinding was significantly transformed. Smartphones provide accurate detection and interpretation of users’ movements and orientation in real-time by using their inbuilt sensor-based systems, such as the Global Positioning System (GPS), an accelerometer, a gyroscope, and a magnetometer. Nevertheless, GPS is not a perfect system—mobile maps can be drawn or labeled erroneously, so even if the above sensors perform correctly, an incorrect map can still be shown [
4]. Aside from occasional map-making errors encountered in mobile maps, two key issues emerge when wayfinding with these systems, namely, (i) the keyhole problem [
5,
6] and (ii) problems with automated systems.
The keyhole problem refers to when elements of interface design restrict the viewing of information on-screen—hindering interface usability. Web browser windows that show cropped portions of images that require a user to resize or scroll the page to display the entire image are an example of this problem. Another example is that some text fields on a website can only accommodate a small section of long words that exceed the field size. In the context of mobile maps, when viewing a map of a large geographic area on a small mobile device, only a small section of the larger map is readily viewable. That is, the user can only perceive a ‘keyhole’ view of the map. This is shown in
Figure 1 below.
There are three interrelated aspects to this problem with mobile maps: (i) the small size and the unstable nature of the map presented on-screen [
11]; (ii) the high interaction cost of manipulating the interface to view the map proper; and the related (iii) view-change cost, where shifts between global and local views need to be understood [
12].
Map instability occurs due to the GPS-based mobile maps moving to correspond with the user’s position or viewing direction and the zoom-in, zoom-out aspect of mobile maps. Interaction cost refers to the number of screen-based interactions a person has to complete in order to complete a task within an interface. A high interaction cost complicates the task [
13]. It is possible that contemporary mobile maps may have a lower interaction cost than earlier versions, such as what was used in earlier studies (e.g., [
9,
11]). Before route directions zoomed-in to upcoming turns in turn-by-turn mode, users needed to zoom in and out of the map much more frequently to increase the details of the maps.
The second key problem when using mobile maps for wayfinding is relying on the automated route directions. Automation is often described as a labor-saving process that can extend or aid human capability. However, automation can also be a “thinking–saving” process where users implicitly trust a system that may fail—with users lacking the ability to deal with the situation. In the case of automated mobile maps, should they malfunction, the wayfinder may not be aware of the current location and might become disoriented or lost. However, when using a paper map to plan a route, the user engages in several cognitive processes to make sense of the map in relation to the surroundings. Lobben [
14] outlines several theories underlying those processes. Three of those are of interest here, namely map rotation, self-location, and visualization, all of which highlight the differences between using paper maps and mobile maps. With map rotation, cartographic tradition prescribes that a “north-up” orientation be used when creating paper maps. This means that in order to relate the map to the environment, the map is rotated, either physically or mentally, to align with the surroundings [
15]. This transforms the map from being allocentric, where the spatial relations between objects in a region are described, to egocentric, where the spatial relations between the map user and objects are described [
16]. Self-location occurs when the map user is able to locate their current location on the map; this process happens at the commencement of a wayfinding task. The physical surroundings, landmarks in particular, are matched to their location on the map, enabling orientation to occur. This is a problem-solving activity [
14]. Visualization, an ongoing cognitive process when wayfinding, refers to another form of map and environment interaction whereby the map’s two-dimensional representation of the surroundings is matched to the three-dimensional real-world environment. For Lobben [
14], unlike matching the environment to the map, as with self-location, visualization involves matching the map to the environment while moving through it. The “mental representations [thus] formed and the patterns people see are closely linked to expectations they bring to a given situation” [
17] (p. 101). Matching the map to the surroundings gives the map-user the ability to predict upcoming spatial configurations in the real world, which is continually repeated as wayfinding continues.
Automation is defined as “the execution by a machine agent (usually a computer) of a function previously carried out by a human” [
18] (p. 231). When using mobile maps, two theoretical spatial procedures are irrelevant: map rotation and self-location are handled by the automated system [
14]. This means that the cognitive effort of paper map rotation is not required as the mobile map system provides a heads-up map, also known as a track-up map. At the simple touch of a button, the map region located directly in front of the user is displayed, which may or may not be north [
19]. The GPS automatically locates the user, which removes the need for self-location. The third of Lobben’s [
14] spatial processes, namely visualization, still occurs with mobile maps. Visualization matches the map to the surroundings as the user moves through the region. However, the amount of visualization provided via a mobile map is probably less than what occurs with a paper map. The small size of the map on mobile phones reduces one’s ability effectively to match the map to the surroundings [
9,
11], which can result in the so-called keyhole problem.
Landmarks present an opportunity for use in the constrained interface of mobile maps as they have been proven to aid our understanding of routes and the surrounding environment [
20,
21,
22,
23]. Research indicates that landmarks are vital environmental elements used when wayfinding. According to some researchers, we use landmarks when we follow a route, in particular at intersections [
24], as landmarks can usefully provide orientation in unfamiliar environments [
20]. We also tend to use them when giving verbal route instructions [
21,
22] and when drawing sketch maps of a route [
23].
Currently, landmarks do not feature in digital mapping systems. With contemporary mobile maps, it appears that technology may exist for depicting landmarks on screen. Here WeGo Map Version 2017), in particular, appear to be the most advanced in this regard. As
Figure 2 shows, the turn-by-turn instructions in Here WeGo Maps use 3D renderings of buildings, and while these buildings may be landmarks in a general sense (i.e., they are prominent buildings/landmarks within cities), they are not used as decision points or confirmation points in route instructions. At present, no technology exists that successfully identifies landmarks based on salience or appropriateness for route instructions and integrates them into a commercially available mapping system [
25,
26].
Deakin [
1] investigated if landmarks presented in a street map would enhance the available spatial information and aid route following. To address this, participants were assigned to one of three participant groups, all of which used a conventional street map. Twenty salient landmarks, all located at intersections, were used. Various geometric symbols depicting the fifteen landmark types (i.e., post office, library, and park) were added to the map given to one participant group; sketches depicting the landmarks were added to the second group’s map, and no supplemental landmarks were added to the map given to the third group. Deakin’s participants interacted with a slide show of the route where they were to choose the correct direction at each intersection to reach the same destination across town—differing from the real-world wayfinding scenario of this research. Participants could refer to their supplied map. The number of errors made and the speed of the participant in making decisions at each decision point were noted.
The two participant groups whose maps included landmarks made significantly fewer turn errors than the no-landmark group, but their decision-making was significantly slower. Deakin [
1] interpreted this as suggesting that the landmarks aided the decision-making process, in particular at complex intersections, as participants claimed that the more realistic sketches helped most at these points. However, it was observed that the users of the symbol landmarks had to refer to a key to understand what each symbol meant, slowing their response, and interpreting some of the sketches may also have slowed progress. In this study, color photographs of landmarks were used as they clearly depict their subject matter, needing no interpretation of the images.
The landmarks selected in this research adhered to a model of landmark salience where conspicuity and prominent location were key selection criteria [
27]—similar to the approach used by Deakin [
1]. Research that investigates wayfinding with landmarks without linking them to concepts of landmark salience (e.g., [
28]) creates uncertainty over the ‘landmarks’ term, leaving it open to subjective interpretation.
4. Results
In this study, all analyses of results were tested for violation of assumptions underlying the Analysis of Variance (ANOVA), the formal analysis techniques applied. These assumptions are as follows: (i) the data do not contain significant outliers, as determined by assessment of a boxplot; (ii) the data shall be normally distributed, as tested by the Shapiro–Wilk test; and (iii) homogeneity of variances exists in the data, as determined by Levene’s test of equality of variances. Efforts were made, when required, to tweak the data to match these assumptions via data transformations. Should these transformations have little effect and parametric modes of analysis are not valid, non-parametric modes, such as a Mann–Whitney U test, were used. Only violations of these assumptions have been reported. Four result categories are provided: (i) Phase 1, the route following task; (ii) Phase 2, the map drawing task; (iii) Phase 3, the landmark locating task; and (iv) Phase 4, correlations between participant task-performance and SBSOD scores.
4.1. Phase 1: Route Following Task
To test hypothesis 1 predicting that the LM Group would outperform the GM Group in regard to accurately following the route, the mean number of errors was calculated for the LM and GM participant groups (LM: M = 0.8, SD = 1.03; GM: M = 1.1, SD = 0.74). Clearly, these figures were too low to warrant a formal error analysis, and the differences between the participant groups were minimal. Hypothesis 1 was therefore refuted.
For the PM Group, failure to adhere to the route described to the researcher prior to embarking on the walk resulted in the errors recorded. In regard to errors, five participants in the LM Group, two in the GM Group, and eight in the PM Group made no errors at all. Only one hesitation was recorded in the PM Group, with no hesitations recorded with the other groups. Seven of the ten PM Group participants chose the simplest route to the destination with only two turns. Regarding the walking times, two outliers were found in the LM Group’s data, as assessed by inspection of a boxplot. These two scores were then adjusted to be slightly above the next highest and below the next lowest scores, a transformation process described by Laerd Statistics [
30]. All data were normally distributed; however, the assumption of homogeneity of variances was violated, as assessed by Levene’s test for equality of variances (
p = 0.017). Therefore, a one-way Welch ANOVA was computed on these means. The walking times ranged from 18:28 min to 26:50 min, with an almost identical mean travel time for the LM and GM Groups (LM Group: M = 22:05 min, SD = 2:56 min; GM Group: M = 22:44 min, SD = 1:07 min) and a slightly shorter time for the PM Group (M = 21:30 min, SD = 2:26 min). These means are shown in
Figure 11.
There were no statistically significant differences in walking time between the three participant groups (Welch’s F(2, 15.187) = 1.1, p = 0.358). It can therefore be concluded that the type of navigational aid guiding the walk did not significantly influence participants’ walking time.
4.2. Phase 2: Route Drawing Task
Hypothesis 2 predicted that the GM Group would draw the route more accurately than the LM Group on an unlabeled paper map after walking the route. Accuracy was determined by the number of erroneously drawn route segments (between each direction point). As shown in
Figure 12, the mean drawing scores for these groups varied somewhat (LM: M = 2.9, SD = 2.4; GM: M = 0.6, SD = 1.35). Two outliers were found and excluded, and the assumption of normality was violated, as indicated by the Shapiro–Wilk test (
p ≤ 0.05). Consequently, a Mann–Whitney U test was conducted. The descriptive statistics for the Mann–Whitney U test provide an overview of the data related to the number of erroneously drawn route segments. The participants (LM and GM Groups) made an average of 1.22 errors when drawing the route. The errors varied widely, ranging from 0 to 5, with a standard deviation of 1.59, indicating considerable variability in participants’ performance. The mean rank for incorrectly drawn route segments was 12.13 for the LM Group and 7.40 for the GM Group, suggesting that the LM Group generally made more errors in drawing the route than the GM Group. The Mann–Whitney U-value is 19.000, with a Z-value of −2.074. The asymptotic significance (
p-value) is 0.038, which is less than 0.05. This indicates a statistically significant difference between the two groups in terms of the number of incorrectly drawn route segments. The LM Group had a higher median number of errors (2.00) compared to the GM Group, which had no errors (0.00) using an exact sampling distribution for U [
31]. This median comparison reinforces the earlier finding that the LM Group made more errors than the GM Group. The results suggest that participants in the GM Group performed better than those in the LM Group in terms of accurately drawing the route, as indicated by both the mean ranks and the median number of errors, thus supporting hypothesis 2. The r-value is approximately 0.49, indicating a medium effect size.
4.3. Phase 3: Locating Landmarks Task
To test hypothesis 3, which anticipated that the LM Group would correctly identify more landmarks on the unlabeled map compared to the GM and PM Groups, a one-way ANOVA was conducted. The analysis used the number of correctly located landmarks as the dependent variable, with the independent variable being the group category, consisting of three levels: LM, GM, and PM. The mean values and standard deviations for the groups were as follows: LM: M = 7.1, SD = 3.51; GM: M = 2.6, SD = 1.96; PM: M = 1.9, SD = 1.79, as shown in
Figure 13. This indicates that the LM Group had the highest average number of correctly located landmarks, followed by the GM Group, with the PM Group having the lowest average.
The ANOVA revealed a significant difference between the groups (F(2, 27) = 12.343, p < 0.001), with a large effect size (η2 = 0.478), suggesting that group membership accounted for a considerable portion of the variance in landmark placement accuracy. Post hoc tests (Tukey HSD) confirmed that the LM Group was significantly more accurate in locating landmarks than both the GM and PM Groups, while no significant difference was found between the GM and PM Groups.
These findings strongly support hypothesis 3, confirming that the LM Group was significantly better at correctly placing landmarks compared to the other two groups. The results imply that the LM Group’s approach or experience was more effective in landmark recognition and placement on the map. Hypothesis 3 was therefore confirmed.
Next, in order to test hypotheses 4 and 5, the data were divided into on- or off-route landmarks. Dealing with hypothesis 4 first, according to which all participant groups were predicted to locate more on-route landmarks correctly than off-route landmarks.
Figure 14 shows this to be true. (LM: M = 5.5, SD = 2.55; GM: M = 2.0, SD = 1.33; PM: M = 1.1, SD = 0.99). Consistent with the prediction of hypothesis 4, the mixed-design 3 x (2) ANOVA computed for the three participant groups (LM, GM, PM) and landmarks (on-route, off-route) revealed a main effect for landmarks (F (1, 27) = 46.00,
p < 0.001). Hypothesis 4 was therefore supported.
Although not specifically hypothesized, a comparison of the data for the on-and off-route landmarks within each of the three groups yielded a main effect for groups as well (F(2, 27) = 15.56, p < 0.001). In addition, the interaction was also significant (F(2, 27) = 15.56, p < 0.001). To determine the locus of the interaction, simple main effects showed that the data for on-route landmarks differed significantly between all three participant groups (p < 0.001). Significant differences also occurred with the locating of on-route landmarks compared to off-route landmarks (LM: p < 0.001; GM: p = 0.004).
With respect to hypothesis 5 predicting that participants in the LM Group would locate more off-route landmarks correctly than participants in the two other groups,
Figure 14 above suggests that to be true (LM: M = 1.7, SD = 1.88; GM: M = 0.7, SD = 1.06; PM: M = 0.8, SD = 1.03). However, the analysis of those data was not as simple as for the on-route landmarks. First, four statistical outliers emerged in the off-route landmark data assessed by inspection of a boxplot. These were therefore reduced downwards. The data were also not normally distributed, as assessed by Shapiro–Wilk’s test of normality (
p > 0.05), and a data transformation failed to rectify this situation. However, as ANOVAs are generally resilient to deviations from normality [
30], the aforementioned mixed-design ANOVA was still carried out. Homogeneity of variances was violated (
p < 0.05), as assessed by Levene’s test, yet there was homogeneity of covariances (
p = 0.09) as described by Box’s M test. The interaction was highly significant, F(2, 27) = 15.56,
p < 0.001, partial 2 = 0.54. To determine the locus of the interaction, simple main effects were calculated for both factors. Contrary to the prediction of hypothesis 5, the analysis for the off-route landmarks yielded no difference between the three groups (LM, GM, and PM), F(2, 27) = 1.086,
p = 0.352, partial 2 = 0.074,
p = 0.352. Hypothesis 5 was therefore rejected.
The data plot in
Figure 15 shows the number of correct placements of the on-route landmarks by their sequential location in the environment. It suggests the presence of a primary and a recency effect, most pronounced for the LM Group. The primacy effect is the effective recall of a particular stimulus first experienced; the recency effect is the effective recall of a particular stimulus last experienced [
32]. Whereas the data for the other two participant groups also suggest a weak primacy effect, the recency effect is largely absent for the GM and the PM Groups. This suggests that, once walking the route, participants in these two groups relied on their navigational aids to follow the route, with no need to refer to landmarks, unlike the LM Group, who relied upon them exclusively. This is further discussed in the
Section 5.
4.4. Phase 4: The SBSOD Scale
The mean SBSOD scores calculated for the three participant groups were quite similar: (LM: M = 3.81/7, SD = 1.33; GM: M = 4.12/7, SD = 0.90; PM: M = 3.21/7, SD = 1.47). Pearson product–moment correlation coefficients were computed to determine the relationship between SBSOD scores and performance for each participant group in Phase 1, which was the only phase involving wayfinding through environmental space. Examination of the number of route-following errors and hesitations in Phase 1 did not reveal any significant correlations either. One might have expected that the PM Group’s scores could yield a correlation, as map-based conditions in other studies have shown correlations, which is further clarified in the next section.
5. Discussion
Hypothesis 1, predicting that the LM Group would outperform the GM Group in regard to accurate route-following, was clearly not supported: all participants in the LM and GM Groups were able to follow the routes described with relative ease, making few errors and hesitations. Even though the route length and complexity for the LM and GM Groups were modeled on previous research [
2,
9,
10], the route was evidently too easy to allow discrimination in the performance of the two groups. To rectify this, a longer and more complex route should be used in subsequent studies to observe more errors and hesitations when route-following. The PM Group generally chose a simple route with two turns, which was easily followed.
According to the confirmed hypothesis 2, the GM Group was more successful at drawing the route on an unlabeled map than the LM Group. In fact, almost all participants in the GM Group were able to draw their routes 100% accurately, resulting in extremely few errors and also confirming that the route was likely too short and easy to follow. Participants in the LM Group drew wrong turns more frequently, and their drawing errors were mainly connected with the large park encountered along the route where three turns were required to navigate around it. A possible reason for this behavior is that the attention of the LM Group when following the route was focused on landmarks, which were predominantly conspicuous buildings (singular points in the region). But as the park was also a landmark (an area in the region), perhaps too much time was spent focused on the park as a landmark when wayfinding rather than on the routes around the park. This also suggests that large area-based landmarks may not perform as well for wayfinding as more point-like landmarks. It is also interesting that the LM Group were able to draw the route—certainly not as accurately as the GM Group—even though they used no map depicting the route. As the PM Group walked different routes, a direct comparison was not possible, but it can be noted that the common two-turn route selected by this group was drawn with almost 100% accuracy. Avoiding complexity, the simplest route will always be the easiest to draw. Another factor contributing to the high performance of the route drawing may have been the scale of the map used. The route sat comfortably in the center of the map area, meaning that participants did not have to work too hard to locate the route area on the map before drawing the route. It appears the pre-set route was too easy to follow, but the length of the route and the number of turns used are close to the mean number of those attributes in similar wayfinding studies [
2,
9,
10,
33].
Regarding locating landmarks, the LM Group located more total landmarks correctly than participants in the GM and PM groups, confirming hypothesis 3, and all three groups located more on-route landmarks than off-route landmarks, confirming hypothesis 4. This makes sense, as the LM and GM Groups followed the same route. Consistent with traditional memory studies, the primacy effect for the LM Group provides evidence that most attention was paid to the first two on-route landmarks and, to a lesser extent, to the last two on-route landmarks, thereby resulting in a weak recency effect as well. Once the first few items (actual landmarks) had been identified by comparing these with the photographs shown on the navigation aid, such comparisons were all they needed to reassure themselves that they were on the right track. With 16 landmarks thus shown, fewer attentional resources were available, although the recent effect suggests that participants still attempted to commit all landmarks to memory. Alas, only the last two stood out, as these were fresh in the participants’ memories. It is also noteworthy that the first and last on-route landmarks were the largest used in this study, which might have further enhanced both the primacy and the recency effects.
The observation that the LM Group also located landmarks on a map that did not indicate the route per se suggests that those participants acquired more than sheer route knowledge, namely an initial level of survey knowledge. This supports the contention that landmark knowledge also contributes to the development of survey knowledge. It is tempting to interpret the observation that the LM Group both recalled more on-route landmarks and were able to place more of these correctly than participants in the other two groups that they had engaged in active encoding of landmark information while walking the route. As they were shown the landmarks again in Phase 2, the results equally suggest it was a case of cued recall.
It is interesting to note that there was some evidence of a primacy effect but not a recent effect on the performance of the GM Group. One way to account for that could be because their navigation tool did not include images of landmarks. Attention would therefore most likely have been focused on following the map-based route correctly. At the start of the route, some GM Group participants were observed visually matching their maps to the environment, with the first couple of nearby on-route landmarks likely noticed. Once progress was satisfactory, participants could trust the map without paying attention to the surroundings. Regarding the PM Group, as many of these participants generally walked routes away from the main route, the on-route landmarks found along this route were not noticed.
The eight off-route landmarks were all visible in the region, but relatively few of these were correctly located on the maps. It is perhaps not surprising that the four off-route landmarks situated nearest to the main route were also those located correctly most often. They were situated 100–150 m from the route, while three of the other four were highly noticeable tall office towers, and another, less noticeable large building, was situated between approximately 300 and 500 m from the route. This suggests that the further a landmark is located from a given path/route, the less likely it will be detected and subsequently recalled.
The refuted hypothesis 5 claimed that because explicit attention was drawn to landmarks, the LM Group would locate more off-route landmarks than the other groups. One possibility is that the route was so simple and the on-route landmarks so obvious that the LM participants had no need to attend to off-route landmarks. Had the route been more complicated, perhaps participants would have needed to rely more on orientation to help them maintain their bearings and understand the space they were traveling through. This is when off-route landmarks may have been noticed as they aid orientation.
As indicated by their mean SBSOD scores, all participant groups rated themselves around average in regard to their perceived environmental spatial abilities. Overall, the scale is most closely related to tasks that involve movement through environmental space where spatial updating is necessary. Research in which correlations have been found has typically involved orientation/direction pointing, route estimation tasks, and only sometimes route-following, as conducted here. Studies reporting correlations between the SBSOD scale and mobile maps in the outdoor wayfinding context are still very scarce. To the best of the authors knowledge, no research has been published in regard to wayfinding solely with sequential images of properly classified landmarks and the SBSOD scale as used by the LM Group here. As Phase 1 directly involved spatial updating, a correlation was predicted in this phase, in particular with the paper map condition given the paucity of correlations with mobile maps. However, the fact that no correlation was observed in that phase suggests that accurately assessing one’s own spatial ability may not be a straightforward task. The route may also have been too easy to follow to truly test participants’ spatial abilities. Alternatively, repeating the route-walking phase may have increased spatial knowledge of the region, thereby demonstrating a stronger correlation with SBSOD scores. The relatively small sample size of 30 participants may have impacted the ability to find correlations here; however, group sizes of 10 are usually adequate to result in statistically reliable group differences.
5.1. Implications of the Research
The research on using photographs of landmarks as navigational aids for pedestrians has the following significant implications:
5.1.1. Enhanced Wayfinding Strategies
This study demonstrates that incorporating landmark imagery into navigation aids can significantly improve wayfinding performance. Participants who used photographs of landmarks made fewer navigation errors and hesitations compared to those relying on traditional map-based systems. This suggests that landmarks serve as effective cognitive tools that enhance spatial awareness and route-following capabilities, particularly in complex pedestrian environments.
5.1.2. Mobile Mapping Applications
The findings advocate for the integration of landmark-based guidance in mobile mapping applications. Current mobile maps do not use images of landmarks in their interfaces. By incorporating landmarks, mobile maps could provide users with more intuitive and contextually relevant navigation experiences, potentially addressing issues like the “keyhole problem” and reducing reliance on automated systems that may fail.
5.1.3. Cognitive Engagement in Navigation
The research highlights the cognitive processes involved in wayfinding, particularly the differences between using landmark-based navigation and other navigation aids. This study suggests that engaging with landmarks requires users to actively participate in their navigation, fostering a better understanding of their surroundings. This contrasts with the passive reliance on automated directions, which can diminish spatial knowledge and awareness.
5.1.4. Future Research Directions
This study opens avenues for further research into the role of landmarks in navigation. Future investigations could explore travel along longer, more complicated routes, how to best use landmarks in map interfaces, and the potential for augmented reality applications that visualize landmarks in real time.