Cardiff map-based orientation talk
1 like204 views
This document summarizes two studies on how people orient themselves using maps in urban environments. The first study found that people often make errors in orientation when relying on highly visually salient objects that are not clear on the map, ignoring important ground-level cues, or misjudging object distances. The second study found that strong 2D ground cues on maps can improve accuracy, but the presence of a salient 3D landmark can confuse people and reduce accuracy. A process model of map-based orientation is proposed based on these findings. The studies have implications for how to design maps to best support orientation.
Cardiff map-based orientation talk
- 1. Map-based orientation in urban environments David Peebles Department of Behavioural and Social Sciences University of Huddersfield 12 April 2011
- 2. Introduction • The project • Funded by OS (2005/08) • Aim: ACT-R model of map-based orientation • Output: Data and high- level process model • This talk • Define orientation • 2 studies investigating orientation strategies • Implications for map design & understanding orientation • A process model of map-based orientation
- 3. Map-based orientation • Key features: • Urban environment • Using an OS map • Goal: identify which way you are facing on map (i.e., align map and environment) • Aim to discover which: • Cognitive processes and strategies used • Map and environmental cues used • Map and environmental factors influence performance
- 4. An example
- 5. Why is this interesting and useful (particularly to OS)? • Cognitive Science • Complex real-world spatial reasoning task • Involves matching egocentric and allocentric mental representations • Requires interaction between cognitive and perceptual processes • Practical implications — find relative importance of: • 2D geometry (local geographic layout) represented on map • If important then use semi-automated cartographic interventions analysing mapped layout • Add landmarks map in places where orientation hardest • Visual 3D scene • If important then add orienting landmarks to maps, particularly most visually salient in scene (Winter, 2003)
- 6. Orientation in real-world environments • Previous research • Rural landscapes (e.g., Eley, 1988; Pick et al., 1995) • Flying perspective (e.g., Wickens & Prevett, 1995; Gunzelmann et al., 2004). • Single building viewed from outside (Warren et al., 1990) • Inside building or single room (Presson, 1982; Meilinger et al., 2007) • Relevant findings • Consistent evidence that people use prominent landmarks (single features & groupings) to solve task if possible • Less evidence that geometry of scene layout is abstracted. Hermer & Spelke (1994) found stronger effect of geometry than landmarks • Urban environments are complex in terms of 3D shape and so important to know if and how landmarks are used
- 7. Experiment 1 • Stimuli • 40 (+5 practice) scenes and maps of Southampton • OS MasterMap Topography Layer at 1:1250 scale. • Participants • 35 University of Huddersfield students & staff • Measures • Angle and response time
- 8. Experiment 1 results • Response correct if within 15 degrees of true angle in either direction • Performance varied between scenes: • 9 scene/graph pairs eliciting accuracy of less than 30% • 10 pairs accurately processed over 70% of the time • Incorrect responses typically clustered around specific features • Focus of analysis: • Identify factors creating common errors • Two independent coders – only agreement of 65% or above included • Four main causes of error. . .
- 9. Scene Object Salience (SOS) • Very visually salient object in the scene (e.g., tall or distinctive building) that is not unambiguous on the map. • 33.2% of erroneous responses in 10 scenes • People identified salient blue building but did not identify it on map • All erroneous responses identified alternative buildings
- 10. Missed Ground-level Cue (MGC) • Unambiguous ground-level cue in the scene (e.g., traffic calming chicane, lawn or pavement) ignored • 25.1% of erroneous responses in 32 scenes. • People identified road on the map • Did not notice the pattern of pathways on either side of the road
- 11. Misperceived Object Distance • Incorrect response would be accurate if object in scene was nearer or further away • 22.4% of erroneous responses in 22 scenes • E & S pointing red lines below — people identified view as lying between two buildings • Misjudged distance of at least one building from viewpoint
- 12. Left/Right Reversal (LRR) • Incorrect response would be accurate if scene was left/right reversed • 17.2% of erroneous responses in 26 scenes. • N pointing red line below — object configuration is mirror image of what is viewed in the scene
- 13. Experiment 1 — conclusions • Factors creating common errors: • SOS — strong effect of visual salience in scene on object selection • MOD & LRR — matching objects but ignoring (or misperceiving) relative depth or left-right asymmetry • MGC — may be due to map used (more detailed than typical street maps) • Limitations of Experiment 1 • Photographs include obtrusive unmapped objects (e.g., parked cars and trees) • Uncontrolled distractors in scenes, (e.g., salient colours or unusual objects) • Matchable objects hidden from view
- 14. Experiment 1 — conclusions • Factors creating common errors: • SOS — strong effect of visual salience in scene on object selection • MOD & LRR — matching objects but ignoring (or misperceiving) relative depth or left-right asymmetry • MGC — may be due to map used (more detailed than typical street maps) • Experiment 2 • Replicate Exp 1 using simplified scenes: • Reduce salience of scene objects • Presence of salient 3D cue and/or distinctive 2D ground layout cue manipulated. • Collect verbal protocols and eye movements
- 15. Experiment 2 • Stimuli • OS 3D buildings model over OS MasterMap Topography Layer on OS Land-Form PROFILE terrain model • 20 (+5 practice) scene/map pairs used in Exp 1 • Participants • 49 University of Huddersfield students & staff
- 16. Experiment 2 results • Errors • 2D ground cue decreased errors (F(1, 48) = 5.47, p < .05) • 3D cue increased errors (F(1, 48) = 40.35, p < .0001) • Interaction effect: • Performance improvement with 2D cue only occurred in absence of 3D one (F(1, 48) = 5.26, p < .05) • Response times • 2D ground cue slowed responses (F(1, 48) = 9.28, p < .005) • 3D cue slowed responses (F(1, 48) = 29.7, p < .0001) • Interaction effect: • Slowing effect of 2D cue small except where a 3D cue was also present F(1, 48) = 4.37, p < .05)
- 17. Conclusions • Experiment 2 • A strong 2D ground cue can improve accuracy (at a time cost) • Unless there is a salient 3D landmark • This may confuse people as to which cue to use (slowing them down more) • People typically opt to use the 3D cue and make as many errors as when the 3D cue appears on its own • General • Identified sources of orientation errors • People tend to seize on visually salient objects in scene • Even when 2D geometry would be a less ambiguous & error-prone cue
- 18. Conclusions • Experiment 2 • A strong 2D ground cue can improve accuracy (at a time cost) • Unless there is a salient 3D landmark • This may confuse people as to which cue to use (slowing them down more) • People typically opt to use the 3D cue and make as many errors as when the 3D cue appears on its own • Recommendations • Best cartographic innovation to help people would be independent of 2D geometry • Provide symbols at points that can be instantly and unambiguously matched to most salient landmarks
- 19. A process model of map-based orientation • A distillation of three approaches: • Warren, Rossano and Wear (1990) • Pick, Heinrichs, Montello, Smith, Sullivan, & Thompson (1995) • Gunzelmann (2008); Gunzelmann & Anderson (2006); Gunzelmann, Anderson, & Douglass (2004) • Key assumptions: • Select features from scene & map (affected by visual salience) • Produce one or more hypotheses • Test hypotheses • Sample additional map/scene features • Attempt to match until sufficient evidence accumulated to decide
- 20. A process model of map-based orientation Match Attempt to identify the selected feature or configuration on the other representation Match found? Testing current hypothesis? Select Select a feature or configuration for matching Scan Scan scene/map for candidate features or configurations to use for mapping Hypothesis supported? Sufficient evidence to make decision? Make decision Hypothesize Generate hypothesis about viewpoint direction Yes No No Yes Yes No Yes No Cognitive factors underlying variation in performance Environmental factors underlying variation in performance Focus initial reconnaissance on scene Organise features into configurations Mental imagery/rotation ability Generate and evaluate multiple hypotheses Compare and test hypotheses using disconfirmation procedure Salience of feature or configuration Scene/map allignment Depth cues Response criteria Speed/accuracy tradeoff
- 21. Things are changing rapidly
- 22. Reference • Davies, C. & Peebles, D. (2010). Spaces or scenes: Map-based orientation in urban environments. Spatial Cognition and Computation, 10, 135–156.