How the Geometry of Space Controls Visual Attention during Spatial Decision Making

How the Geometry of Space controls Visual Attention during Spatial Decision Making Jan M. Wiener (jan.wiener@cognition.uni-freiburg.de) Christoph H¨olscher (christoph.hoelscher@cognition.uni-freiburg.de) Simon B uchner (simon.buechner@cognition.uni-freiburg.de) Lars Konieczny (lars.konieczny@cognition.uni-freiburg.de) Center for Cognitive Science, Freiburg University, Friedrichstr. 50, D-79098 Freiburg, Germany Abstract In this paper we present an eye-tracking experiment investi- gating the control of visual attention during spatial decision making. Participants were presented with screenshots taken at different choice points in a large complex virtual indoor en- vironment. Each screenshot depicted two movement options. Participants had to decide between them in order to search for an object that was hidden in the environment. We demonstrate (1.) that participants reliably chose the movement option that featured the longest line of sight, (2.) a robust gaze bias to- wards the eventually chosen movement option, and (3.) using a bottom-up description that captures aspects of the geometry of the sceneries depicted, we were able to predict participants’ fixation behavior. Taken together, results from this study shed light onto the control of visual attention during navigation and wayfinding. Keywords: visual attention; wayfinding; navigation; gaze be- havior; spatial cognition; spatial perception. Introduction What controls visual attention when navigating through space? In the context of navigation, eye-tracking studies so far primarily investigated the role of gaze for the control of locomotory or steering behavior (Grasso, Prevost, Ivanenko, & Berthoz, 1998; Hollands, Patla, & Vickers, 2002; Wilkie & Wann, 2003). Wayfinding, however, also includes pro- cesses such as encoding and retrieving information from spa- tial memory, path planning, and spatial decision making at choice points (c.f. Montello, 2001). So far, very few, if any, studies made use of eye-tracking techniques to investi- gate such higher level cognitive processes involved in navi- gation and wayfinding. For example, which information do navigators attend to and process when deciding between path alternatives? And, how does gaze behavior relate to spatial decision making at all? To approach these questions we pre- sented participants with images of choice points and asked them to decide between two movement options while record- ing their eye-movements. In non-spatial contexts, gaze behavior has been shown to reflect preferences in visual decision tasks (Glaholt & Rein- gold, in press). In two alternative forced choice paradigms in which participants have to judge attractiveness of faces, for example, gaze probability is initially distributed equally between alternatives. Only briefly before the decision, gaze gradually shifts towards the eventually chosen stimulus (Shi- mojo, Simion, Shimojo, & Scheier, 2003; Simion & Shimojo, 2007). It is an open question whether similar effects can also be observed in spatial decision making such as path choice behavior. The features people attend to when inspecting images of scenes have been investigated in numerous studies revealing both, bottom-up (stimulus derived) as well as of top-down (e.g., task) influences (for an overview see Henderson, 2003). Already in the 60s, Yarbus (1967) demonstrated influences of the task on the control of visual attention: participants’ gaze patterns when inspecting the same drawing systemati- cally differed when asked to judge the ages of people depicted or when asked to estimate their material circumstances. The most widely used bottom-up approach is that of saliency maps (Itti & Koch, 2000, 2001). A saliency map is a representa- tion of the stimulus in which the strength of different fea- tures (color, intensity, orientation) are coded. Several studies demonstrated that saliency maps are useful predictors of early fixations, particularly when viewing natural complex scenes (e.g., Foulsham & Underwood, 2008). It is important to stress that bottom-up approaches usu- ally do not explicitly account for the fact that images or pic- tures are two-dimensional projections of three-dimensional scenes. In other words, the geometrical properties of the scenes depicted in the images are not necessarily captured or highlighted by, for example, saliency maps. For naviga- tion and wayfinding, however, the interpretation and under- standing of the depicted three dimensional structure may be inevitable. This opens up intriguing questions: Is it possible to predict gaze behavior by analyzing geometrical properties of the sceneries depicted if the viewer is solving a navigation task? If so, can the analysis of gaze behavior be used to infer the strategies and heuristics underlying different navigation or wayfinding tasks? And, which kind of description systems of spatial form and structure captures properties of space that are relevant for the control of visual attention? Promising candidates are isovists or viewshed polygons (Benedikt, 1979), which both describe the visible area from the perspective of the observer. Isovists are essentially depth profiles and several quantitative descriptors such as the visi- ble area, the length of the perimeter, the number of vertices, etc., can be derived that reflect local physical properties of the corresponding space. Moreover, isovists have been shown to capture properties of the geometry of environments that are relevant for experience of the corresponding space and locomotion within the space (Wiener et al., 2007; Franz & Wiener, 2008). The specific research questions for this study were as fol- lows: