**4. Methods of eye-tracking data visualization**

Results of eye-tracking measurements are presented as a text file containing a timestamp and a number of specifications describing coordinates of the point of regard, the pupil size, the angle of the eye position etc. First of all, it is necessary to classify the data, with a specified algorithm, as fixations and saccades (see chapter Eye movements and algorithms of their detection). Then, the data are even visualized in a suitable way, or can be statistically analysed.

There are several basic methods of the eye-tracking data visualization. Holmqvist et al. [20] present the main techniques of gaze data visualization as follows: ScanPath (GazePlot), Attention (Heat) maps and the AOI (Area of Interest) Analysis.

Following chapters will present different possibilities of visualization and eye-tracking measurements on concrete examples from cartography. In this way, methods of eye-tracking will be shown. All examples are based on source data, which are results of several authentic experiments on cartography rules evaluation.

All presented case studies were performed using the SMI RED 250 remote device with the sampling frequency of 120 Hz. Eye position was measured every 8ms.

Respondents were chosen from university students of Geoinformatics and Cartography and also from other studying fields which are not related to cartography.

## **4.1. ScanPath**

106 Cartography – A Tool for Spatial Analysis

In a different study the same authors [32] deal with a generic evaluation approach combining theory and data-driven methods based on sequence similarity analysis. The approach systematically studies users' visual interaction strategies when using highly interactive map interfaces. The result was that the participants generally follow a sequence

Another application of eye-tracking in cartography appears in the study of Opach and Nossum [33] where authors have explored the suitability of eye-tracking on two different semistatic and traditional cartographic animations of temperature and weather. Contrary to the author's previous web based experiment, analysis of the eye-tracking data revealed that the viewing behaviour were surprisingly similar. Three of the metrics used (fixation counts, observation length and time to first fixation) indicated very similar viewing strategies and

Fuhrman, Tamir and Komogortsev [34] have dealt with an assumption that threedimensional topographic maps provide more effective route planning, navigation, orientation, and way-nding results than traditional two-dimensional representations. The eye-tracking metrics analysis indicates with a high statistical level of condence that three-

Popelka and Brychtova [35] used eye-tracking together with questionnaire investigation for evaluation user's attitudes toward interactive methods of virtual geovisualisation of changes in the city built-up area. Five approaches of visualization were assessed - textual description of changes, comparison of historical and recent pictures or photos, overlaying historical maps over the orthophoto, enhanced visualization of historical map in large scale using the

Technologies and methods of eye-tracking have not yet been fully utilized in cartography, even though the possibilities are wide. Cartographic research in the field of eye-tracking currently focuses explicitly on improving the user quality of maps. Future potential expansion of eye-tracking technology can be seen in the activity of the International Cartographic Association, especially the Commission on Use and User Issues [36] and the newly established

Results of eye-tracking measurements are presented as a text file containing a timestamp and a number of specifications describing coordinates of the point of regard, the pupil size, the angle of the eye position etc. First of all, it is necessary to classify the data, with a specified algorithm, as fixations and saccades (see chapter Eye movements and algorithms of their detection). Then, the data are even visualized in a suitable way, or can be

There are several basic methods of the eye-tracking data visualization. Holmqvist et al. [20] present the main techniques of gaze data visualization as follows: ScanPath (GazePlot),

third dimension and photorealistic 3D models of the same area in different ages.

Commission on Cognitive Issues in Geographic Information Visualization [37].

**4. Methods of eye-tracking data visualization** 

Attention (Heat) maps and the AOI (Area of Interest) Analysis.

statistically analysed.

that agrees with the hypothetical sequence representing user's strategies.

behaviour during viewing different kind of cartography animations.

dimensional holographic maps enable more efficient route planning.

ScanPath is defined as a route of oculomotor events through space within a certain timespan [20]. Thanks to ScanPath, it is possible to display raw data as well as calculated fixations and saccades. Circles of different sizes represent fixations (their radius corresponds with their length) and lines which connect the circles represent saccades [38].

When a larger amount of data is displayed, this method becomes restraining. Overlapping parts of individual fixations cause that it is not possible to identify their number visually. Figure 3 shows an example of a ScanPath. Respondents were asked to find the highest peak on one of the maps. The aim of this experiment was to compare two types of visualization perspective 3D display and a classical orthogonal map supplemented with the shading. Raw data are displayed on the left of the picture, fixations and saccades on the right. From both pictures, it is evident that this particular respondent preferred the three-dimensional visualization. His answer is displayed by a red dot which represents the mouse-click.

**Figure 3.** ScanPath showing raw data (left) and fixation and saccades (right)

#### **4.2. ScanPath comparison**

There is a great need for robust and general method for ScanPath comparison existing in many fields of eye-tracking research [39]. Privitera and Stark [40] introduced ScanPath comparison based on string editing. Fixations are replaced with characters standing for the AOI's they hit and the ScanPath is represented as character string. It is one of the rst methods comparing not only the loci of fixations, but also their order. The principle of this method is the transformation of two-dimensional data (X, Y coordinates of fixations) to onedimensional data (character string). Each ScanPath is recorded as a string of letters where each letter corresponds to the area of a current fixation location.

Advanced Map Optimalization Based on Eye-Tracking 109

measurement or calibration deflection. By means of HeatMaps, it is possible to get an

HeatMaps produced by SMI software BeGaze are created in two steps. The software first scales each pixel in proportion to the durations of all fixations landing on it. Typically, this results in a very sparse "Fixation Hit Map", since only a small proportion of the pixels have been "hit". In next step, the hit map is convolved with a Gaussian kernel with certain width.

Areas of Interest (AOI) are regions in the stimulus which the researcher is interested in. The most important AOI metric is the dwell time defined as one visit in AOI, from entry to the exit. The dwell has its own duration, the starting point, the ending point, dispersion etc. In several ways, it is similar to a fixation, but it is of much larger entity, both in space and time [20]. It is also possible to follow the order in which the respondent looked at particular areas

In cartography the Areas of Interest analysis can be used advantageously. AOI analyses are based on evaluation of concrete parts of a map (legend, scale, title, specific phenomena in the map, etc.). When evaluating influence of a composition on the map reading, the application of AOI is very useful. By indicating and evaluating particular compositional elements as AOI, several characteristics can be find out - e.g. for how long the respondent

The results can be visualized with use of the Sequence Chart, which displays observed areas in different colours on a timeline. Figure 6 shows results of an experiment, whose objective

overview of possible measurement inaccuracies and so to adapt the Area of Interest.

A wider kernel gives a smoother, less pointy appearance to the attention map [20].

**Figure 5.** HeatMap created from fixations of seven respondents

or the transition, the movement from one AOI to another etc.

was observing the given area, in which order he visited them etc.

**4.4. Area of Interest** 

Two or more character strings are then compared and their similarity is measured. Dened by an optimization algorithm, string editing assigns a unit value to three different character operations: deletion, insertion, and substitution. Characters are then manipulated in order to transform one string into another and character manipulation values are tabulated [41]. String edit algorithm determines the number of operations needed to transform one sequence to another - the operations being insertions, deletions and substitutions. The calculated metric will be a measure of how different two sequences are. This method uses the Levenshtein algorithm to produce a string-edit distance between each sequence [42]. Example of the Levenshtein distance measure is in the figure 4. Character string comparison methods are widely used in bioinformatics to align DNA and protein sequences.


**Figure 4.** Result of the Levenshtein distance measure between group of eleven geoinformatics while observing a map. Gridded AOI with grid of 5\*5 cells was used

Depending on specific tasks, it is important to distinguish between gridded AOI and semantic AOI approaches. When using the Gridded AOI, stimulus is split into areas of equal size (rectangles) with no relation to semantics. The second approach, Semantic AOI uses the Area of Interest which corresponds to specific areas in stimuli. In cartography, Semantic AOI is generally more advantageous because it corresponds to map composition elements like title, legend, etc.

#### **4.3. Attention maps**

Attention maps, also called HeatMaps, are used for visualization of quantitative characteristics of the user's gaze. Thanks to attention maps, it is possible to identify to which area user pay attention and which are rather neglected. In eye-tracking, HeatMaps enable the creation of a brief summary of areas which are in the spotlight and so they are needed to be analyzed more thoroughly. Except of the function of visualization, they might be used as a background for AOI creation. When plotting AOI around a small object, fixations of some of the respondents could be noted outside the created AOI because of inaccurate measurement or calibration deflection. By means of HeatMaps, it is possible to get an overview of possible measurement inaccuracies and so to adapt the Area of Interest.

HeatMaps produced by SMI software BeGaze are created in two steps. The software first scales each pixel in proportion to the durations of all fixations landing on it. Typically, this results in a very sparse "Fixation Hit Map", since only a small proportion of the pixels have been "hit". In next step, the hit map is convolved with a Gaussian kernel with certain width. A wider kernel gives a smoother, less pointy appearance to the attention map [20].

**Figure 5.** HeatMap created from fixations of seven respondents

#### **4.4. Area of Interest**

108 Cartography – A Tool for Spatial Analysis

method is the transformation of two-dimensional data (X, Y coordinates of fixations) to onedimensional data (character string). Each ScanPath is recorded as a string of letters where

Two or more character strings are then compared and their similarity is measured. Dened by an optimization algorithm, string editing assigns a unit value to three different character operations: deletion, insertion, and substitution. Characters are then manipulated in order to transform one string into another and character manipulation values are tabulated [41]. String edit algorithm determines the number of operations needed to transform one sequence to another - the operations being insertions, deletions and substitutions. The calculated metric will be a measure of how different two sequences are. This method uses the Levenshtein algorithm to produce a string-edit distance between each sequence [42]. Example of the Levenshtein distance measure is in the figure 4. Character string comparison

methods are widely used in bioinformatics to align DNA and protein sequences.

**Figure 4.** Result of the Levenshtein distance measure between group of eleven geoinformatics while

Depending on specific tasks, it is important to distinguish between gridded AOI and semantic AOI approaches. When using the Gridded AOI, stimulus is split into areas of equal size (rectangles) with no relation to semantics. The second approach, Semantic AOI uses the Area of Interest which corresponds to specific areas in stimuli. In cartography, Semantic AOI is generally more advantageous because it corresponds to map composition elements like title, legend, etc.

Attention maps, also called HeatMaps, are used for visualization of quantitative characteristics of the user's gaze. Thanks to attention maps, it is possible to identify to which area user pay attention and which are rather neglected. In eye-tracking, HeatMaps enable the creation of a brief summary of areas which are in the spotlight and so they are needed to be analyzed more thoroughly. Except of the function of visualization, they might be used as a background for AOI creation. When plotting AOI around a small object, fixations of some of the respondents could be noted outside the created AOI because of inaccurate

observing a map. Gridded AOI with grid of 5\*5 cells was used

**4.3. Attention maps** 

each letter corresponds to the area of a current fixation location.

Areas of Interest (AOI) are regions in the stimulus which the researcher is interested in. The most important AOI metric is the dwell time defined as one visit in AOI, from entry to the exit. The dwell has its own duration, the starting point, the ending point, dispersion etc. In several ways, it is similar to a fixation, but it is of much larger entity, both in space and time [20]. It is also possible to follow the order in which the respondent looked at particular areas or the transition, the movement from one AOI to another etc.

In cartography the Areas of Interest analysis can be used advantageously. AOI analyses are based on evaluation of concrete parts of a map (legend, scale, title, specific phenomena in the map, etc.). When evaluating influence of a composition on the map reading, the application of AOI is very useful. By indicating and evaluating particular compositional elements as AOI, several characteristics can be find out - e.g. for how long the respondent was observing the given area, in which order he visited them etc.

The results can be visualized with use of the Sequence Chart, which displays observed areas in different colours on a timeline. Figure 6 shows results of an experiment, whose objective

was to reveal differences in reading of a simple map by a group of students of cartography and cartography amateurs. Three different map compositions are displayed in figure 6. Maps (presented in the first row of figure 6) were projected in 5s intervals during which the respondents have to observe maps without answering any question. The second raw in the Figure 6 shows a sequence chart for a group of students of Geoinformatics and cartography, who have attended several cartography courses. The last row represents data given by cartographic amateurs, students of psychology, zoology etc.

Advanced Map Optimalization Based on Eye-Tracking 111

In statistics, the result is called statistically significant when it is unlikely to have occurred by chance alone, according to a pre-determined threshold probability, the significance level. Critical tests of this kind may be called tests of significance. When such tests are available, we can say whether the second sample is/ is not significantly different from the first one

From a more detailed evaluation of different types of basic eye movements (fixations and saccades), or gaze data metrics such as dwell time, it is possible to deduce a series of

Different composition perception with two different groups of map users (experienced cartographers and non-cartographers) was verified by the testing of measured results by a two sample t-test, which is a method of mathematical statistics making possible to verify the

Differences of the mean dwell time of two groups of users were tested on particular Areas of Interest of the map list - title of the map, the map, the legend, the imprint and secondary maps. A zero hypothesis H0 was tested: mean values of particular choices are the same.

Main map -2,2189 33,374 0,0334 0,05 2375,58 3080,11 Rejecting H0

heading 3,7546 51,501 0,0004 0,05 631,27 230,78 Rejecting H0

map 2 0,0963 36,715 0,9238 0,05 387,84 376,47 Fail to reject H0

map 1 -0,1874 40,321 0,8523 0,05 260,59 284,08 Fail to reject H0

Masthead 1,1999 36,275 0,2380 0,05 127,27 57,57 Fail to reject H0

legend 0,2211 45,702 0,8260 0,05 99,50 84,33 Fail to reject H0

By comparing the mean dwell time in particular AOI it is evident that both groups spent most time on the main map field. Extremely long time was noticed at students of cartography in AOI covering the map title. The same AOI was on the 4th position with non-

The t-test result disproved the zero-hypothesis saying that the values of the mean dwell time were the same with AOI map heading and the main map field. The visit rate of the main map field was significantly higher with non-cartographers. On the contrary, the map

**Table 1.** Results of two sample t - test of dwell time on particular compositional map elements of two

carto [ms]

mean of non-

carto [ms] statement

null hypothesis that the means of two normally distributed populations are equal.

numeral metrics suitable for other statistical testing.

Concrete results are illustrated in table 1.

t df p-value alpha mean of

[44].

Map

Additive

Additive

Map

different user groups.

cartographers.

Each stimulus was preceded by a short cross used to locate beginning of all trajectories at the same place (in the middle of the picture). That is why the AOI representing the map field is always pictured in first 500 ms. After this time; most geoinformatics students automatically read the title of the map, or rather noted fixations representing it in AOI. Cartographic amateurs did not do so. It is evident especially in the first column, where the stimulus was the "ideal" map composition [43]. In following columns, the composition was not in accordance with cartographic rules. Despite this fact, students of geoinformatics were trying to find the title of the map.

**Figure 6.** Sequence chart visualization. Sample data of the Esri were used for the creation of stimulus maps

The Sequence Chart is illustrative and easy to interpret, but it is necessary to evaluate the data by statistical approach.

The objective of the statistical hypotheses testing is to evaluate the data gained from experiments and the suitability of the purpose given before the testing. Statistical hypothesis is a certain purpose about the distribution of accidental quantities of a basic file.

In statistics, the result is called statistically significant when it is unlikely to have occurred by chance alone, according to a pre-determined threshold probability, the significance level. Critical tests of this kind may be called tests of significance. When such tests are available, we can say whether the second sample is/ is not significantly different from the first one [44].

110 Cartography – A Tool for Spatial Analysis

trying to find the title of the map.

maps

data by statistical approach.

was to reveal differences in reading of a simple map by a group of students of cartography and cartography amateurs. Three different map compositions are displayed in figure 6. Maps (presented in the first row of figure 6) were projected in 5s intervals during which the respondents have to observe maps without answering any question. The second raw in the Figure 6 shows a sequence chart for a group of students of Geoinformatics and cartography, who have attended several cartography courses. The last row represents data given by

Each stimulus was preceded by a short cross used to locate beginning of all trajectories at the same place (in the middle of the picture). That is why the AOI representing the map field is always pictured in first 500 ms. After this time; most geoinformatics students automatically read the title of the map, or rather noted fixations representing it in AOI. Cartographic amateurs did not do so. It is evident especially in the first column, where the stimulus was the "ideal" map composition [43]. In following columns, the composition was not in accordance with cartographic rules. Despite this fact, students of geoinformatics were

**Figure 6.** Sequence chart visualization. Sample data of the Esri were used for the creation of stimulus

The Sequence Chart is illustrative and easy to interpret, but it is necessary to evaluate the

The objective of the statistical hypotheses testing is to evaluate the data gained from experiments and the suitability of the purpose given before the testing. Statistical hypothesis

is a certain purpose about the distribution of accidental quantities of a basic file.

cartographic amateurs, students of psychology, zoology etc.

From a more detailed evaluation of different types of basic eye movements (fixations and saccades), or gaze data metrics such as dwell time, it is possible to deduce a series of numeral metrics suitable for other statistical testing.

Different composition perception with two different groups of map users (experienced cartographers and non-cartographers) was verified by the testing of measured results by a two sample t-test, which is a method of mathematical statistics making possible to verify the null hypothesis that the means of two normally distributed populations are equal.

Differences of the mean dwell time of two groups of users were tested on particular Areas of Interest of the map list - title of the map, the map, the legend, the imprint and secondary maps. A zero hypothesis H0 was tested: mean values of particular choices are the same. Concrete results are illustrated in table 1.


**Table 1.** Results of two sample t - test of dwell time on particular compositional map elements of two different user groups.

By comparing the mean dwell time in particular AOI it is evident that both groups spent most time on the main map field. Extremely long time was noticed at students of cartography in AOI covering the map title. The same AOI was on the 4th position with noncartographers.

The t-test result disproved the zero-hypothesis saying that the values of the mean dwell time were the same with AOI map heading and the main map field. The visit rate of the main map field was significantly higher with non-cartographers. On the contrary, the map

title visit rate was higher with cartographers. There was no difference of the dwell time statistically proved with other observed AOI.

Advanced Map Optimalization Based on Eye-Tracking 113

followed almost the same gaze trajectories. One of the first fixations of both respondents are localised in the legend. Then, the respondents tried to answer the given question and started

**Figure 7.** Time-space data displayed by means of Space-Time-Cube.

**5. Conclusion and prospects** 

**Figure 8.** Use of Space-Time-Cube visualization for investigation of ScanPaths from two respondents

Up to now, technologies and methods of eye-tracking in cartography were not fully utilized despite their great possibilities in cartography. A cartographical research in the field of eye-

to explore the map.
