How to effectively represent spatial information on handheld mobile devices is a key question given the increasing use of personal digital assistants (PDAs) and cellphones concurrent with the development of location-based services. The mobile use of digital maps on small displays presents new capabilities and challenges that differ from using paper maps in a mobile setting or viewing digital maps on a desktop computer. This research addresses these issues through a pilot study that evaluates maps on a mobile device used for a field-based task.
Department of Geography
University of California, Santa Barbara
Map representations at two levels of generalization were compared by analyzing subject performance in an on-foot route-following task with a handheld computer used as a navigation aid. In examining subject time and accuracy as well as interaction with the mobile device during the task, the results carry implications for map design for small, mobile displays and identify factors that affect the use of maps while moving. Maps are and will increasingly be used on small displays in mobile contexts for a variety of purposes and in many different environments. The requirements and preferences of mobile users, as well as how these maps are used in different contexts, must be understood in order to inform effective designs.
Understanding what makes maps effective for small displays and how users interact with digital maps on handheld devices such as personal digital assistants (PDAs) and cell phones is critical to map design for mobile computing. Mobile GIS displays allow users to zoom in and out, to add or remove data layers, to change the appearance of the display, and even to modify the dataset itself. New capabilities resulting from mobility, ranging from the ability to interact with data to taking complex datasets into the field, give greater flexibility to maps and mapmakers. How people use these capabilities while actually engaged in mobile activities, whether as a sightseeing tourist or data-collecting researcher, needs to be understood in order to inform map design for these devices.
This research reports on a pilot study that examined subjects’ performance during an onfoot navigation task with a handheld computer. Subjects followed routes marked on maps at two different levels of generalization, an aerial photograph and a classified, simplified version of the aerial photo. One focus of the experiment was to evaluate the level of map generalization with regard to three dependent variables: time to route completion, amount of map browsing, and accuracy in following the route. The statistical analysis indicates that the generalized map performed better than the aerial photograph, with significantly shorter time to route completion, and the use of fewer zoom levels and fewer zoom changes.
A second focus considered subjects’ spatial abilities, familiarity with the study area, and experience with maps and mobile technology. Examining these factors with subjects’ performance begins to address questions regarding user behavior with maps on handheld devices. What do patterns of map browsing reveal about how users behave when disoriented? Are there consistencies in the types and amounts of errors that people make, and how is error related to spatial ability? When characteristics of subjects, especially spatial ability, are included in the analysis, the conclusion that the generalized map was more effective is called into question. Overall, the results point to a variety of factors that affect the use of a mobile map, as well as a large variation in the way individuals interact with a digital map on a handheld device.
Representations starting at the least-generalized end of a representation spectrum were chosen for this study: a photorealistic image and a manually-created generalized map. A color aerial photograph at a scale of 1:12,000 was scanned to digital format and used as one display condition. Taken 14 months prior to the study, it portrays the actual environment in terms of detail and color, with no cartographic design applied. The generalized map is a classified and simplified version of the aerial photograph, created by manually tracing all readilydistinguishable landscape features in a GIS, then color-coding the polygons according to feature type: buildings, sidewalks, grass and other vegetation, trees, paved roads, sand and water. The objective was to evaluate two representations that were equivalent in feature information, with the only difference in the level of generalization being a reduction of detail and classification of features.
The use of the aerial photograph for this study represents a baseline, or even the “worst case” scenario in terms of map design, since there has been no design applied. Comparing the results with a generalized version of the same dataset, in which basic techniques of classification of feature types and simplification of detail have been used to create the map, the differences are attributable to the level of generalization, rather than other design factors, such as labels or symbology, that would be present in a map created by a cartographer. In addition, with aerial photography and high-resolution satellite imagery widely and increasingly available to the public, it is a convenient type of dataset to use as a map or backdrop image for a variety of purposes. In terms of its value as a realistic representation style, it has been argued (Bishop 1994) that the general public is more comfortable viewing realistic maps than generalized maps, since interpreting a realistic, hence familiar, scene is more intuitive than interpreting an abstract scene, especially to an inexperienced map user. The National Park Service has been moving towards more realistic maps since the 1980s, for example, incorporating texture and relief detail from aerial photography in order to make more user friendly maps for park visitors (Patterson 2002). However, depending on the purpose of the map, and in this case for use in a mobile, navigational context, the high level of detail in the aerial photograph carries the potential to overwhelm the user or make the map impossible to read on a small display. The shadows and small distortions inevitable in aerial photography may be confusing. The power of a generalized map, designed to a specific purpose and simplified in terms of any of a number of techniques, is that it can focus attention on information that is relevant to the map purpose (Visvalingam 1994). A generalized map may reduce cognitive load in terms of the user visually processing the image; however, reconciling the abstract representation to the real environment may introduce another burden (Bishop 1994).
2. Related Research
This study complements continuing research on spatial information delivery for mobile devices, and is unique in considering controlled variations of map generalization in a field-based task. It is the first part of a series of studies to systematically test carefully controlled variations of representations to determine what makes effective mobile cartography, and why.
Dynamic, digital maps are key applications for mobile devices, especially for providing navigation assistance to non-expert users, or assisting scientists and others who work with spatial data in the field. A recent special issue of Cartography and Geographic Information Science dedicated to mobile mapping and GIS identified a research agenda for mobile GIS, encompassing the areas of infrastructure, data, and user issues (Clarke 2004). A major research area is navigation assistance and location-based services. Prototype navigation aids, both handheld and wearable varieties, continue to be developed and tested. Commercial products, such as in-car navigation systems, are already available to consumers. Research has not been able to keep pace with the technology, however, even for digital maps for standard-size computer monitors; there are no cartographic design guidelines yet for digital maps as there are for traditional paper ones (Meng 2003). There is a growing body of research investigating the variety of spatial information presentation available for handheld and wearable mobile devices: visual maps in 2D and 3D, text/audio descriptions, schematic diagrams, ground-view photography or video, or combinations of these. A review of these systems is beyond the scope of this paper (see Urquhart, Cartwright et al. 2003 for an overview of the more major projects), but many of these studies are concerned with how well a navigation system works as a whole, on a technological or usability level, such as which representation type or modality is more effective, rather than trying to determine why one representation method is better in light of how users interact with the information.
Map generalization research for small display is faced with the technological challenge associated with the limited screen space of mobile devices, and is driven by the need for automatic methods of creating representations that can adapt to the user’s context (Edwardes, Burghardt et al. 2003), or can change scales and levels of detail in real-time (Hampe and Sester 2002). One of the primary problems of automatic generalization is devising a way to represent only the information that is relevant to the user at a particular time (Agrawala and Stolte 2001). The question of what that relevant information is remains to be determined.
A conceptual framework for approaching what information is important to represent on a mobile map has been developed by a number of researchers based on the context of the user and the mobile device. Since different types of information are necessary for different purposes and user activities, and given that the capabilities of GIS and digital images allow for dynamic maps that can potentially custom- tailor the amount and type of information displayed to individual users, the context of mobile map use is a starting point for designing effective spatial representations (Reichenbacher 2004). Nivala and Sarjakoski (2003) discuss mobile map context from the broad, mobile computing context categories of Computing, User, Physical, Time and History, defining more specific categories related to maps, encompassing hardware and infrastructure, and how, where and by whom the map on the device is used. They emphasize a need for research to determine which context factors are most important to incorporate for designing a map, and how exactly to do it (Nivala and Sarjakoski 2003).
Considering navigation systems specifically, Hampe and Elias (2004) focus the idea of context on the user, his navigation purpose, and his situation: individual characteristics of spatial skills, experience and familiarity with the area; whether he is moving by car, bicycle or foot; navigation style preferences; and characteristics of the situation, such as the time of travel, season, traffic conditions, etc. These factors are taken into consideration to determine the best way to present navigation information for a given context, such as which landmarks are going to be relevant, and which presentation modality fits with the attention and interaction limitations of the user (Hampe and Elias 2004). The information that this framework depends on, especially as regards how users’ abilities and experience affect the way they use mobile maps, still needs to be determined.
An extensive framework of context for mobile cartography has been developed by Reichenbacher (2004), and is largely concerned with the adaptive nature of maps on mobile devices. This approach emphasizes that the information content for the map (such as areal extent and level of detail) and the information visualization (scale, generalization, symbolization, etc.) are categories in which elements can be presented in a specific and optimal way for the user and his situation (Reichenbacher 2003). Reichenbacher (2004) reviews current approaches and outlines several specific research directions for mobile cartography.
The methodology and results from this pilot study inform these contextual frameworks by assessing patterns of behavior with subjects of different spatial abilities and experience using maps of varying representation in a controlled experiment. These results are specific to the type of environment of the study area and the activity of route- following, and would likely be different if the experiment were replicated in a different type of area, such as a downtown city center or a forest, with a different sized study area, and if subjects were finding their way to a destination point rather than following a given route. Systematically considering maps in all contexts is a necessary step towards a more complete understanding of how people interact with maps while mobile. This knowledge will inform generalization techniques, such as in determining just what level of detail is necessary, or to what extent features can be aggregated or simplified to fit on the display and still be useful.
In order to evaluate map generalization for handheld computer displays in a mobile context, an experiment was designed to have subjects use digital maps to complete a navigation task. Research subjects were 16 graduate students from different departments at the University of California, Santa Barbara (UCSB), 10 males and 6 females, ages ranging from 20 to 37. Their task entailed walking along a route displayed on a map on a tablet PC, using the map as a navigation aid. Subjects were instructed to complete the route as accurately as possible and as quickly as possible, but walking at their normal pace. Subjects did the task with both display conditions, following a different route each time. The two display conditions are shown in Figure 1. Each route was the same length, 0.74 km, contained the same number of turns (19), and covered similar-sized, non-overlapping areas of the UCSB campus (Figure 1). Routes and map order were systematically varied among participants to avoid confounds from any differences in the two routes or from practice effects. Prior to starting the task, subjects received training with the tablet PC and completed short practice routes with each map type, to get familiar with the interface, display and task instructions. The navigation task was designed so that subjects would need to interact with the device continually, referring to the map in determining where to walk.
The hardware platform was a ruggedized tablet PC (Figure 2a), and a Java application displayed the map image in a 6 x 6 cm window, the size of the display dimensions of a typical PDA-sized handheld computer (Figure 2b). While this research is concerned with the limitations of small displays, using a tablet PC, which runs a Windows XP operating system, allowed more flexibility and control over the experimental design and data collection than a PDA would have. A pan frame surrounded the four sides of the image, with incremental zoom in and zoom out buttons below the image window (Figure 3). All buttons were selected by touching the screen with a finger. Figure 4 shows zoom levels 3, 4 and 5 for the aerial photograph, with spatial resolutions of (a) 1 meter, (b) 0.5 meter, and (c) 0.25 meter, respectively.
The study area was the UCSB main campus, although the area extent that the subjects walked during the task was only a small portion of the entire area, approximately 0.07 square kilometers. A view of the full extent of campus was available to subjects for purposes of orientation, and subjects began the task with the image at its full extent. Maps were oriented on the device the conventional north- up, but subjects were free to physically rotate the device to rotate the map. Indeed, all subjects rotated the map during the navigation task according to the direction in which they were heading, consistent with findings in other research that users prefer to use a map oriented to their direction of travel (Warren and Scott 1993; Bornträger, Cheverst et al. 2003). Such physical rotation of the map, or of one’s body, in order to line up the orientation of the map to the real world reduces the amount of mental rotation required of the subject to reconcile the map with the real environment (Aretz and Wickens 1992). To begin the task, subjects were taken to the start point of the route, given the tablet computer with the map zoomed all the way out, oriented towards north to match the map orientation, and shown the start point of the route on the map. Before beginning to walk the route, subjects zoomed and panned to orient themselves according to their preference, pushed the button labeled “start”, and began walking. Upon reaching the end of the route, subjects pushed the “stop” button.
It should be noted that although there is a GPS integrated into the tablet PC, no location information appeared on the map. This first study was concerned with how subjects performed with just the map and the route, requiring subjects to maintain their location themselves. An important follow-up study will be to replicate the experiment with location information from the GPS, to understand the effects of such assistance.
One goal of the study was to evaluate the maps with regard to time efficiency, amount of map browsing in the form of zooming, and accuracy.
Time efficiency considered the total amount of time it took each subject to complete the route for each display condition. Faster completion of the navigation task would indicate a more effective map. It was hypothesized that the generalized condition would result in faster time to route completion. The amount of map browsing was defined as the amount of zooming done by a subject in completing the task. Pushing buttons takes time and requires user attention, and keeping track of the changing display during zooming requires the subject to remember previous views, as opposed to viewing a static display. Also, the action of zooming suggests that the subject needs additional information; therefore, less zooming interactions would suggest a more effective map, one with enough information in the current display. It was thought that the generalized map would require less zooming than the aerial photograph map, given the comparatively lesser amount of detail and higher contrast among feature types. The ability to complete the task accurately was seen as an important factor to consider. A map type that contributes to user error is not an effective map. For the task, subjects were instructed to follow the route marked on the map as closely as possible, and were told that the route would not necessarily follow sidewalks or paths. Accuracy was measured by the number of errors subjects made along the route with respect to features. Walking around the wrong side of a tree, for example, or walking on sidewalk where the route indicates to walk on grass were counted as errors. The aerial photograph was expected to result in fewer errors than the generalized map, since it contained more details and visual cues than the generalized map.
A second focus of the experiment was to consider subjects’ performance as related to their spatial abilities, previous experience with maps and mobile devices, and familiarity with the study area. It has been shown (for example, Hegarty, Richardson et al. 2002; Prestopnik and Roskos-Ewoldsen 2000; Sholl, Acacio et al. 2000) that individual differences have a significant impact on performance with environmental spatial tasks of navigation and wayfinding. It was hypothesized that these characteristics would affect subjects’ performance on the task, and therefore would show significant correlations with time to route completion, amount of map browsing, and accuracy.
Data was collected in the form of computer logs, observations by the researcher, and through questionnaires completed by the subjects. The mobile device logged all operations to a file, recording each action of zoom in, zoom out, and move up, down, left, and right, along with pixel coordinates of the image center, current zoom level, and time to the nearest second. During the task, subjects were followed at a short distance by a researcher, whose function was ostensibly to assist with any technical problems or inquisitive passers-by, but who also recorded observations on where and when subjects stopped or made errors. A questionnaire completed by each subject prior to the task used a 5-point scale to collect self-report data on general familiarity with the UCSB campus, experience with using maps to navigate, experience with mobile computers such as PDAs or video games, and asked a series of questions to assess sense of direction spatial abilities. Spatial ability, in terms of environmental orientation, was measured using the Santa Barbara Sense of Direction Scale, a self-report survey that provides a reliable, quantified assessment of the type of spatial abilities associated with locating oneself in an environment and maintaining orientation during movement through an area (Hegarty, Richardson et al. 2002).
After completing the navigation task with both conditions, subjects completed a questionnaire to evaluate and compare both maps. Questions on a 5-point scale asked subjects about their familiarity with the areas covered by the routes, to rate and comment on the difficulty of using the device in general, and for each map, to rate the level of detail, indicate which feature types were most and least useful for orientation, which characteristics, of shape, texture, size or color, were most and least helpful in identifying features, and whether subjects felt they used the pan and zoom extensively or not very much. In addition, open-ended questions asked which of the two maps was easier to use and why, and had subjects comment on the most difficult thing about following the route. These questions were designed to acquire information about user perceptions and preferences about the maps and the mobile device, as well as to identify subject behavior or strategies that could be linked to the quantitative data from the log file.
While subject interaction with the map in terms of zooming varied quite a bit among individual subjects, there were significant differences in behavior from one condition to the other. Almost half of the subjects used only one or two zoom levels during the tasks, while the rest used three or more different zoom levels. Between conditions, subjects changed zoom level an average of twice per route for the generalized map, but an average of nine times with the aerial photograph. Individual behavior varied considerably, with some subjects using one zoom level for the entire task and others using 6 out of the possible 7.
These results suggest that the higher level of detail in the aerial photograph compared to the generalized map prompted subjects to use a higher level of zoom to distinguish features and determine where to navigate. With the generalized map, at a certain point it would be evident to the subject that zooming closer would provide no further information, thereby reducing the inclination to zoom further.
No significant correlations were found between amount of zooming and spatial abilities, familiarity with the study area, experience with maps or experience with mobile devices. However, the statistical results of the map browsing data when taken as a whole may mask patterns in zooming behavior in particular situations, such as zooming to maintain position or get more detail on an area versus when a subject is disoriented or lost. Therefore subjects’ behavior in terms of map zooming when they made errors was considered. With the capability to zoom in or out for a different view, did subjects take advantage of this or not? Where subjects made errors, it seems that in most cases they recognized a problem or an ambiguity; many people would stop and study the map before continuing on. For this analysis, making an error, and especially stopping beforehand, is taken to indicate disorientation in the subject. In looking at how subjects utilized the zoom function when they were disoriented, eight subjects never zoomed at the time that they made an error, five subjects regularly did zoom when disoriented, and two subjects sometimes did and sometimes did not.
This behavior seems to be related to prior experience with handheld computers. Subjects’ report of their previous experience with handheld computers such as PDAs when compared to their zooming behavior shows that those with more experience tended to zoom when disoriented. Those with less experience may have not remembered the zooming capability at that moment, being preoccup ied with reorienting themselves. Or, they may have preferred not to zoom for some reason. Perhaps they were more used to traditional static maps, or found a changing view more confusing than helpful. This is an area that needs to be investigated more closely; all subjects understood the zooming function and tried it out during the practice session, but, at critical times when their attention was intently focused on the task of reorienting themselves, not everyone used the function. If novice users do not or cannot take advantage of a function that can assist them at critical times, that tool is not effective.
Overall, the statistical analysis points to the generalized map as being more effective than the aerial photograph for the route- following task. However, when individual differences of the subjects are considered, the determination of which map was more effective depends on the subject. For subjects with poorer environmental spatial abilities, the aerial photograph condition seemed to be more effective, resulting in about the same or even faster time to route completion than the generalized map.
Considered another way, both representations worked: all subjects were able to successfully complete the task with both map types. While it seems clear that a well-designed map would be preferable for use on a mobile device over a photograph or other remotely sensed image, the advantage of the latter is that it can be more up-to-date than a designed map, and requires minimal processing. Despite ongoing efforts to automate processes, generalization requires the skills of a cartographer, time, and of course, resources. For maps used in a mobile context, especially for navigation and wayfinding, having accurate, up-to-date information is a priority. Thus, what is the tradeoff between current, relatively raw imagery, and an older but carefully designed map? In this study, one of the major locations of error was a place where a wall feature did not appear on the map. It was concluded that this missing information in the map was the cause of confusion and error for a majority of subjects. Missing or erroneous information in a map can be the result of recent construction in an environment or a natural event, beyond any purposeful omission or aggregation of features as dictated by the cartographer or the map scale. Research by Casakin, Barkowsky et al. (2000) concluded similarly that distorted or oversimplified depictions of path and intersection features in a map cause confusion and error in wayfinding. The result from patterns of errors in this study, taken together with the conclusion that there was no overall difference in accuracy between the map conditions, suggests that a more current map, even with a lot of detail, would be preferable to an outdated but well-designed map.
This study demonstrated that considering the locations and types of errors subjects made was important in the analysis in that patterns were revealed that were masked by the statistical analysis of the data taken as a whole. The causes of and influences on the errors described here are subject to interpretation, but provide a foundation for further research to test specific situations in which there is ambiguity in the map due to obscured features, missing information, poor image contrast at key intersections, or an inappropriate amount of detail. The patterns of how subjects of different spatial ability and familiarity with the area behave can help in anticipating problem areas during map creation, and can inform design decisions to try to prevent confusion for the user.
As a pilot study, this research points to directions for the focus of continuing studies, and provides a starting point from which to test additional levels of generalization in map representation for mobile devices. As noted in the previous section, priorities for further research on mobile maps include the relationship of environmental spatial ability to level of map generalization, the effect of familiarity with the environment on performance, and the influences on zooming preferences for novice users and those with more practice using maps on mobile devices. Additionally, investigating the influence of a location indicator from GPS is an important next step. Subjects commented that keeping track of where they were on the route and which direction they were heading was difficult at times. Reducing the burden for the user of determining his current location by incorporating GPS information supports more effective movement through an environment (Suomela, Roimela et al. 2003). In that case, what bearing does location information have on the map representation? Can the map then be generalized to a minimalist schematic, or do users still need a certain amount of detail for context? How much detail?
Beyond these questions, the roles of previous experience with maps, and continuing practice with digital maps on a mobile device must be considered. Interaction behavior of a novice user would likely change as that user becomes more familiar with the device and discovers his preferences for zooming and the type of information he uses for navigation and wayfinding, which in turn would have an effect on the information that he would need displayed in the map. This research focused on a route- following task designed to require subjects to continually interact with the map as they made their way through an environment. Different types of user activity were not considered, but assessing map representations for use in additional activity contexts is necessary for a more complete understanding of what makes maps effective for handheld mobile devices. Alternative mobile devices, such as wearable computers with head- mounted displays, provide another mechanism of spatial information delivery, and maps designed for handheld systems would likely need to be different for a heads-up display.
Finally, while the experimental design proved successful overall for achieving the goals of the study, the research was not without challenges. Visibility of the computer display in varying illumination conditions, especially bright sunlight, was a challenge, and one that has been noted by others (such as Kray, Elting et al. 2003). As mobile devices continue to advance technolo gically, this issue may be mitigated. The other major limitation was the logistics and time constraints associated with field studies. Routes had to be carefully designed to be in areas for which the aerial photograph was accurate, long enough to require subjects to use panning and zooming, but not so long that the task would take an excessive amount of time. Each subject spent about one hour completing the experiment, including the practice routes, two conditions, and questionnaires. Rescaling the amount of subjects for a bigger study is a significant endeavor; therefore it is crucial that continuing experiments be carefully designed and prioritized. Some of the questions for further study raised here can be initially addressed through lab-based studies, but ultimately, since the mobile context is so different from an indoor, relatively static setting, field testing is necessary to gain a true understanding of mobile map use.
This pilot study sought to investigate how people use digital maps on small devices while engaged in mobile activities, by comparing two variations of a map representation, and considering characteristics such as environmental spatial ability, familiarity with the area, and previous experience with maps and handheld devices. While the results are contextualized for this type of study area and population, the relationships and patterns that emerge demonstrate the importance of considering individual subjects’ characteristics when approaching mobile map design. Future studies will further elaborate and verify these relationships, and consider additional variations of context. Ultimately, an understanding of how people interact with digital maps while moving will bring effective spatial representations to mobile devices, whether they are being used for navigation, data collection, or other field-based activities.
No Comments »
Leave a comment
You must be logged in to post a comment.