Spatial Predictors of Eye Movement in a Gallery Setting

Full text

Spatial Predictors of Eye Movement in a Gallery Setting
Jakub Krukar and Ruth Conroy Dalton ?
Northumbria University, Newcastle-upon-Tyne, United Kingdom
Abstract. The impact of space on our behaviour and cognition is not yet fully understood.
The problem is particularly interesting in the context of art galleries, where the spatial context
of artefacts is probably the most impactful curatorial tool available, which greatly contributes
to the visitors’ final experience. Space Syntax - a set of methods for quantitative description
of spatial environments - was used to extract various properties of objects’ locations from a
gallery’s layout. Among other variables, eye-tracking measures were obtained from partici-
pants freely exploring the space. The results of our analyses show that the spatial arrangement
of objects is highly correlated with the number of glimpses occurring at given location. The
implications of this are relevant to researchers in real life eye-tracking studies interested in
setting up highly controllable experimental spatial environments.
1 Introduction and the Context
The role of museum curators is to arrange objects into exhibitions within the possibilities and
constraints offered by the available spaces. These spaces have often been designed, prior to any
curatorial endeavours, by a studio of architects with an entirely different set of goals. Each group
has its own priorities and assumptions of the effect the exhibition space might have on the eventual
visitor. Hence, the intersection of curation, building design and visitor behaviour remains an area
awaiting greater understanding. This project attempts to explain this relation by investigating the
influence of spatial layout of art galleries1on human memory and attention for exhibitions.
To what do we attend [1] and what do we ‘get out of it’ [2] are perhaps the most crucial
factors in determining a usable and effective museum space. In psychological terms this problem
is tightly linked to the notions of visual attention and memory. The connection between these two
processes, even though established [3, 4], remains only partially explained. Moreover, as it is widely
acknowledged, the findings from laboratory studies might not always be generalisable to real-life
activities [5]. And since mobile eye-tracking research is in growing demand [6] it becomes critical to
establish cognitive results (and causes) of our oculomotor behaviour in real world environments. All
of these aspects are at the centre of interest among museum curators whose task it is to facilitate
the above mentioned processes through the manipulation of space and artefacts it contains.
Landmark research seem to be a fertile starting point for these investigations as they focus on
describing what type of objects draw our attention in real life spaces and remain in the accessible
memory after we change our position [7]. As it has been suggested, our attentional resources are
not necessarily drawn by the salience of the object but more likely by its spatial position [8, 9].
This general conclusion is in line with eye-tracking studies of 2D scenes, where the concept of the
salience map [10] does not provide the satisfactory explanation to where we look [5]. Concurrently,
the fact that objects’ salience might be given an overestimated importance has been suggested in
?The authors wish to thank Sheep Dalton for his time-demanding support with the equipment.
1We refer to ‘art galleries’ as the subset of all ‘museums’, studies of which are also often relevant here.

the museum context [11]. Yet, it still would come as a surprising finding that what shapes our
experience of a museum exhibition is more the ‘spatial container’ than the contents.
The environment of art galleries, compared to other everyday spaces we experience, is fairly
deprived of visual distractors. Paintings hanging on the walls are intuitive and obvious candidates
to fixate one’s attention on - they are the reason we come to art galleries in the first place. As the
result of this visual simplicity, the arrangement and configurational relationships between smaller
parts of the larger space become the most powerful curatorial tool available [12]. This tool has
been already shown to distinguish the way knowledge is transmitted by the exhibits [13], as well
as facilitate the movement [14] and engagement [15] of the visitors. Experimental approach like the
one we here present can enrich the long tradition of visitor studies conducted with observational
methods and post-visit questionnaires [1]. Eye-tracking can reveal much more precise insights into
what the visitors of an art gallery really saw. This, as we attempt to show, is largely affected by
how the space was arranged for them.
2 Method
Thirty-two participants (13 female, mean age = 30.75, SD = 11.73; one removed from the data
analysis due to previous exposure to the space) divided into 2 groups individually explored a
private art gallery wearing Tobii Glasses mobile eye-tracking device. Later, they undertook two
unanticipated memory tests: a computer recognition task (where reaction times were recorded) and
a test similar to the ‘Tour Integration’ task described in [16] - participants were given a printed
layout of the visited space together with miniatures of the pictures seen inside and were asked to
place them onto their recollected location.
Fourteen pictures (Fig. 1) were hung in the gallery, on the locations presented in Fig. 2. All
participants saw the same pictures, but the sequence (i.e. placement at the specific location) was
randomised for each person. Knowing what was the location of each picture in the gallery for the
given participant allowed gaze movement and the subsequent memory test results to be coded in
two ways, separately: as location-oriented variables (Which locations attracted more fixations;
which locations’ content was memorised better?), and as picture-oriented variables (Which of the
pictures generated more fixations; which pictures were memorised better?).
Fig. 1. Pictures used in the study

Fig. 2. Locations of the pictures in both conditions (Cond. 1 has a single picture located towards the centre
of a wall, Cond. 2 has multiple pictures occuring at the same wall).
To investigate the relation between space and oculomotor behaviour in this paper, we focus
mainly on the location-oriented variables. Analysed eye-tracking measures involved (mean values
for each location were obtained):
1. total dwell time (defined as the sum of all fixation times within the picture’s boundaries; linked
previously to the meaning of the viewed area, its informativeness and interest [3], as well as the
memory for object’s position [4]);
2. normalised total dwell time as the percentage of the whole time spent inside the gallery;
3. number of dwells on each location (defined as the discrete number of times eyes’ scanpath entered
each object’s boundaries and at least one fixation occurred; related to object’s informativeness
and the confidence levels in object recognition [3]);
4. normalised number of dwells for each location as the percentage of all dwells.
Space Syntax measures used are derived from the concept of the visibility graph [17] and
Benedikt’s formulation of the isovist [18]. For the Visibility Graph Analysis (VGA) [17] the layout
of our gallery is divided into a small grid, and each of the grid’s cells (equivalent to ∼0.5 m2in
real space) can be treated as a node of a highly connected graph. For example, the number of other
nodes which can be ‘seen’ by any single node (given the walls’ constraints) contributes to that
node’s connectivity value. DepthMap 10.14.00b [19] was used for spatial analysis.
Another important concept used in the analysis is the isovist together with its properties [18].
An isovist is a 2D polygon bounding the area visible from a particular point (the centre point of each
picture’s location in this case). Purely geometrical characteristics of this polygon (such as its area)
can be extracted. Point 2nd moment for example reflects the dispersion of isovist’s perimeter [18],
and APR (area/perimeter ratio) is a commonly accepted measure of jaggedness (although various
transformations exist, e.g. [20]). We explain the crucial measures while discussing the results, but
please refer to the original papers [17–19] for a detailed description.

3 Results
Four two-way ANOVAs with picture and condition as independent variables were conducted: one
for each of the four eye-tracking dependent variables. Three of them (except normalised number of
dwells ) showed a significant effect of the condition at at least p<.05.
Table 1 and Table 2 present Pearson’s Product-Moment Correlation values calculated separately
for Cond. 1 (locations x101–x114; N= 14) and Cond. 2 (locations x201–x214; N= 14) as well as
jointly for both conditions (N= 28).
Table 1. Pearson’s Product-Moment Correlation values for Dwell Time; *p<0.05; **p<0.01; ***p<0.001
Mean Total Dwell Time Mean Normalised Total Dwell Time
Cond.1+2 Cond.1 Cond.2 Cond.1+2 Cond.1 Cond.2
Other Objects Within Isovist 0.14 0 0.69** 0.09 -0.22 0.51
Visual Clustering Coefficient -0.03 0.02 -0.24 -0.1 0.09 -0.38
Visual Control 0.42* 0.65* 0.42 0.48* 0.63* 0.45
Visual Controllability 0.47* 0.49 0.66** 0.54** 0.51 0.64*
Visual Entropy -0.40* -0.56* -0.07 -0.23 -0.39 0.04
Visual Integration 0.57** 0.62* 0.64* 0.57** 0.58* 0.56*
Visual Mean Depth -0.58** -0.64* -0.67** -0.60*** -0.60* -0.65*
Connectivity 0.52** 0.55* 0.67** 0.56** 0.55* 0.63*
Isovist Area 0.52** 0.56* 0.67** 0.56** 0.56* 0.63*
VCA Area 0.43* 0.4 0.48 0.58** 0.62* 0.52
Isovist Compactness -0.39* -0.37 -0.41 -0.38* -0.26 -0.47
Isovist Max. Radial 0.46* 0.32 0.57* 0.54** 0.39 0.64*
Isovist Min. Radial 0.36 0.56* 0.49 0.47* 0.5 0.62*
Isovist Occlusivity 0.46* 0.49 0.5 0.48** 0.44 0.53
Isovist Perimeter 0.49** 0.51 0.58* 0.52** 0.45 0.59*
Point First Moment 0.54** 0.58* 0.67** 0.60*** 0.60* 0.64*
Point Second Moment 0.56** 0.58* 0.67** 0.61*** 0.62* 0.62*
Area/Perimeter Ratio (APR) 0.37* 0.21 0.71** 0.45* 0.3 0.65*
4 Discussion and Conclusion
The results of ANOVA suggest the importance of spatial configuration in differentiating the visual
experience of a gallery visit.
Those spatial variables which generated similar correlation values for both conditions are the
most promising candidates for spatial predictors of eye-motion, as they seem to remain stable in
different spatial settings. Connectivity and Isovist Area are essentially the same concepts calculated
in two different ways [17,18]. Their high correlation with number of dwells can be explained with
regard to pure probability: assuming any random path through the studied environment and the
biological constrains of our visual system, objects whose isovist covers a larger proportion of the
area, are more likely to fall in sight. This event does not, however, determine the total length of
such gaze and hence the correlation values with total dwell time are lower. Other factors (such as

Table 2. Pearson’s Product-Moment Correlation values for Number of Dwells; *p<0.05; **p<0.01;
***p<0.001
Mean No. of Dwells Mean Normalised No. of Dwells
Cond.1+2 Cond.1 Cond.2 Cond.1+2 Cond.1 Cond.2
Other Objects Within Isovist 0.26 0.06 0.69** 0.41* -0.06 0.71**
Visual Clustering Coefficient -0.01 0.12 -0.22 -0.11 0.12 -0.28
Visual Control 0.71*** 0.78*** 0.82*** 0.79*** 0.76** 0.86***
Visual Controllability 0.77*** 0.77** 0.93*** 0.81*** 0.69** 0.91***
Visual Entropy -0.34 -0.47 -0.1 -0.27 -0.48 -0.17
Visual Integration 0.80*** 0.82*** 0.84*** 0.78*** 0.74** 0.85***
Visual Mean Depth -0.75*** -0.84*** -0.78*** -0.74*** -0.78** -0.81***
Connectivity 0.80*** 0.81*** 0.91*** 0.81*** 0.73** 0.90***
Isovist Area 0.80*** 0.82*** 0.91*** 0.82*** 0.74** 0.90***
VCA Area 0.84*** 0.84*** 0.90*** 0.82*** 0.83*** 0.86***
Isovist Compactness -0.44* -0.28 -0.56* -0.45* -0.3 -0.61*
Isovist Max. Radial 0.60*** 0.65* 0.55* 0.59*** 0.68** 0.61*
Isovist Min. Radial 0.25 0.14 0.53 0.31 0.09 0.47
Isovist Occlusivity 0.59*** 0.56* 0.65* 0.61*** 0.57* 0.69**
Isovist Perimeter 0.68*** 0.62* 0.79*** 0.70*** 0.59* 0.82***
Point First Moment 0.85*** 0.90*** 0.91*** 0.86*** 0.84*** 0.90***
Point Second Moment 0.87*** 0.92*** 0.88*** 0.86*** 0.89*** 0.87***
Area/Perimeter Ratio (APR) 0.60*** 0.53* 0.76** 0.60*** 0.45 0.74**
personal preference) might be at play in case of this variable. Taking into account Salience Rating
of the investigated pictures (assessed in our separate study, not reported in this paper) suggests
that objects’ salience is not one of them.
In addition to Isovist Area, its version restricted to a 60 degree visibility cone was calculated,
similarly to previous suggestions in spatial analyses of gallery settings [21]. This measure - here
called Visibility Catchment Area (VCA) - attempts to capture the field of meaningful visibility and
ignore the size of the area from which a picture cannot be seriously visually investigated or at least
distinguished from other pictures. VCA generated a bit higher correlation values from Isovist Area,
which is an empirical confirmation of its aforementioned theoretical advantage.
The highest correlation values for number of dwells was, however, reached by Point Second Mo-
ment (see definition above). This suggests, that it is not necessarily the pure size of the isovist,
but also its shape that accounts for our visual behaviour. Although it must be noted that, in both
of our layouts, Isovist Area was highly correlated with Point Second Moment (r= .93), and this
situation is not always true. Since Point Second Moment is a concept much worse described in the
literature, we recommend Isovist Area and it’s derivatives, such as VCA to be used as spatial pre-
dictors of visual behaviour. At the same time, shape of the isovist requires more thought, especially
considering the fact that APR, which is a more commonly used measure of isovist jaggedness did
not generate meaningful results in the case of our study.
Various isovist properties have been recently shown to predict the memorability of displayed
words and images from public displays [22], navigational behaviour [20] and spatial decision making
while viewing 2D scenes [23], so the discussion of the issue is ongoing.
Presented high correlation values between spatial predictors and oculomotor behaviour can have
an important impact on the future research in mobile eye-tracking. Space Syntax can be a feasible

tool helping to control experimental environments or aid curators in their tasks when unintentional
preference given to the spatial presentation of some objects over others is undesired.
References
1. Bitgood, S.: An attention-value model of museum visitors. Technical report, Center for the Advance-
ment of Informal Science Education. (2010)
2. Vergo, P.: The Reticent Object. In Vergo, P., ed.: The New Museology. Reaktion, London (1989) 41–59
3. Holmqvist, K., Nystr¨om, M., Andersson, R., Dewhurst, R., Jarodzka, H., Van de Weijer, J.: Eye
tracking: A comprehensive guide to methods and measures. Oxford University Press, Oxford (2011)
4. Tatler, B.W., Gilchrist, I.D., Land, M.F.: Visual memory for objects in natural scenes: from fixations
to object files. The Quarterly Journal of Experimental Psychology 58A(5) (July 2005) 931–60
5. Land, M.F., Tatler, B.W.: Looking and acting. OUP, Oxford (2009)
6. Tatler, B.W.: On the perception of natural scenes: an introduction to the special issue. Journal of Eye
Movement Research 2(2) (2008) 1: 1–4
7. Caduff, D., Timpf, S.: On the assessment of landmark salience for human navigation. Cognitive
processing 9(4) (December 2008) 249–67
8. Janzen, G.: Memory for object location and route direction in virtual large-scale space. Quarterly
journal of experimental psychology (2006) 59(3) (March 2006) 493–508
9. Miller, J., Carlson, L.: Selecting landmarks in novel environments. Psychonomic bulletin & review
18(1) (February 2011) 184–191
10. Itti, L., Koch, C.: A saliency-based search mechanism for overt and covert shifts of visual attention.
Vision Research 40(1012) (June 2000) 1489–1506
11. Newhouse, V.: Art and the Power of Placement. The Monacelli Press, New York (May 2005)
12. Zamani, P.: Architecture as Curatorial Device. In Koch, D., Marcus, L., Steen, J., eds.: Proceedings
of the 7th International Space Syntax Symposium, Stockholm, KTH (2009) 134:1–12
13. Peponis, J., Hedin, J.: The layout of theories in the Natural History Museum. 9H 3(1982) 21–25
14. Hillier, B., Ma jor, M., Desyllas, J., Karimi, K., Campos, B., Stonor, T.: Tate Gallery, Millbank: a study
of the existing layout and new masterplan proposal. Technical report, UCL (January 1996)
15. Wineman, J., Peponis, J., Conroy Dalton, R.: Exploring, Engaging, Understanding in Museums. Space
Syntax and Spatial Cognition Workshop: Spatial Cognition ’06 (2) (2006) 33–51
16. von St¨ulpnagel, R., Steffens, M.C.: Can Active Navigation Be as Good as Driving? A Comparison of
Spatial Memory in Drivers and Backseat Drivers. Journal of Experimental Psychology-Applied 18(2)
(2012) 162
17. Turner, A., Doxa, M., O’Sullivan, D., Penn, A.: From isovists to visibility graphs: a methodology for
the analysis of architectural space. Environment and Planning B 28(1) (2001) 103–121
18. Benedikt, M.: To take hold of space: isovists and isovist fields. Environment and Planning B 6(1979)
47–65
19. Turner, A.: A program to perform visibility graph analysis. In Peponis, J., Wineman, J., Bafna, S.,
eds.: Proceedings of the 3rd Space Syntax Symposium, Atlanta, University of Michigan (2001) 31.1–9
20. Wiener, J.M., Franz, G., Rossmanith, N., Reichelt, A., Mallot, H.A., B¨ulthoff, H.H.: Isovist analysis
captures properties of space relevant for locomotion and experience. Perception 36(7) (2007) 1066 –
1083
21. Stavroulaki, G., Peponis, J.: The spatial construction of seeing at Castelvecchio. In: Proceedings of the
4th International Space Syntax, London (2003) 66:1–13
22. Dalton, S., Marshall, P., Conroy Dalton, R.: Measuring Environments for Public Displays : A Space
Syntax Approach. In: CHI ’10 Extended Abstracts, New York, ACM (2010) 3841–3846
23. Wiener, J.M., H¨olscher, C., Buechner, S.J., Konieczny, L.: How the Geometry of Space controls Visual
Attention during Spatial Decision Making. In: Annual Meeting of the Cognitive Science Society. (2009)