Evaluating the Effectiveness of Occlusion Reduction Techniques for 3D Virtual Environments

1
Evaluating the Effectiveness of Occlusion
Reduction Techniques for 3D Virtual Environments

• Niklas Elmqvist
• Chalmers University of Technology
• Gothenburg, Sweden
• ltelm_at_cs.chalmers.segt

M. Eduard Tudoreanu University of Arkansas at
Little Rock Little Rock, AR, USA ltmetudoreanu_at_ualr
.edugt
2
The Matchup!
vs.
3
Introduction

• Most meaningful 3D worlds are large
• Too large to be seen from a single viewpoint
• Many objects, many information-carrying entities
• Cluttering and occlusion becomes an issue
• To solve this we can introduce an occlusion
  management technique
• Definition intelligent management of 3D views to
  minimize (or eliminate) inter-object occlusion
• Examples 3D thumbnails, navigational aids,
  distorting space, etc
• For an overview, see Elmqvist and Tsigas 2007
  (IEEE Virtual Reality 2007)

4
Introduction (2)

• Problem there are plenty of techniques, but few
  guidelines on when to use each one
• Our approach comparative study of different
  occlusion management techniques
• Test techniques for different tasks
• Identify the best technique for each task
• Develop a set of design guidelines for when to
  choose different occlusion management strategies
• Method between-subjects user study involving 16
  subjects performing tasks w/ the techniques
• Counting, relation, pattern identification

5
Generalized Fisheye Views

• Aggregation of overview and detail views into one
  single view
• Standard overviewdetail has two separate views
• Components
• One or more foci
• Degree-of-interest (DOI) function
• The level of detail (LOD) of each data point is
  derived using the DOI function and the distance
  from the current focus

New Yorkers View of the World (Saul Steinberg,
1976)
6
Generalized Fisheye Views (2)

• Our implementation is designed for 3D virtual
  environments
• One focus centered on the users input device
• Continuous DOI function based on Euclidean
  distance
• Interest inversely proportional to distance
• i(d) c / (d c) constant c
• Level-of-detail based on two different metrics
• Transparency change transparency of objects
  depending on the interest
• Scale change size depending on interest

7
Generalized Fisheye Views (3)
8
BalloonProbe

• Interactive 3D space distortion technique
• Inflateable force field connected to the users
  input device
• Can be inflated and deflated
• Smooth displacement of objects
• Purpose disambiguation of targets in areas of
  high object congestion
• Different probe geometry for different purposes

9
BalloonProbe (2)

• Two main purposes for using the probe
• Distractor removal get rid of distractors from
  view
• Target separation separate targets to more
  easily access them
• Two main probe geometries
• Sphere continuous and smooth displacement,
  facilitates target separation
• Wedge parting branches, facilitates distractor
  removal
• Our implementation includes both of the above

10
BalloonProbe (3)
11
User Study

• The purpose of the user study was to identify
  strengths and weaknesses of each technique
• Predictions
• BalloonProbe best for local tasks, fisheye for
  global
• Standard navigation (base case) poorest
• Three main hypotheses
• H1 Technique is significant for performance,
  accuracy, distance and rotation
• H2 There is interaction between technique and
  task
• H3 Any occlusion management technique
  outperforms simple 3D navigation on all four
  metrics

12
User Study (2)

• Subjects 16 participants (4 female) 20 to 35
  years old
• Equipment three-sided CAVE environment (front,
  right, and floor)
• Design
• Independent variables
• Within-subjects technique x density x world x
  trial
• Between-subjects task count, pattern, relate
• Dependent variables
• Time, error, distance travelled, total angular
  rotation
• Counter-balanced order of conditions

13
User Study (3)

• Three tasks (from before)
• Count counting the number of yellow cones in the
  world
• Pattern identify the pattern formed by the
  yellow cones in the world (C, R, K, X, or Y
  letters)
• Relate find the third target spying on the
  yellow cone and green box targets
• Density was low or high (100 or 200 objects)
• World was either abstract or office
• Abstract designed to mimic infovis application,
  very simple 3D primitives
• Office indoor 3D environment (akin to an
  architectural walkthrough)

14
Results Time Performance

• Time average for all techniques is 2 min 10 sec
• No techique is significantly faster than all
  others

15
Results Time Performance (2)

• However, not best in any of the individual tasks
• Clearly, different techniques have different
  strengths

16
Results Correctness

• Spherical BalloonProbe is significantly more
  accurate than the other techniques
• No technique is the least accurate method

17
Results Correctness (2)

• Again, breakdown by task shows that different
  techniques do better at different tasks

18
Discussion

• Looking at our hypotheses, we see the following
• H1 Technique is significant for speed, accuracy,
  distance and rotation
• Proved! Technique is a significant effect for all
  of the above
• H2 There is interaction between technique and
  task
• Proved! Different tasks require different
  techniques
• In general, designing an end-all be-all
  interaction technique is known to be very
  difficult (if even possible)
• H3 Any occlusion management technique
  outperforms simple 3D navigation on all four
  metrics
• Disproved! Simple navigation is significantly
  faster than all occlusion management techniques
• It is the least accurate, however...

19
Discussion (2)

• There is clear trade-off between speed and
  accuracy
• Simple 3D navigation with no occlusion management
  is best for quick-and-dirty viewing
• Occlusion management is appropriate when mistakes
  incur a high cost
• Spherical BalloonProbe seem to be the best of the
  occlusion management techniques tested
• Consistency in both speed and accuracy for all
  tasks
• Predictability is a useful property for
  general-purpose interaction techniques

20
Conclusions

• We have presented an empirical study evaluating
  the strengths and weaknesses of different
  occlusion management techniques
• 3D fisheye techniques (scaling and transparency)
• BalloonProbe (spherical and wedge-shaped)
• Standard 3D navigation
• Results indicate a strong task preference
  different task require different techniques
• Standard 3D navigation is fine for quick viewing
• Spherical BalloonProbe affords the best accuracy
• Suitable for high-precision tasks

21
Questions?

• Contact information
• M. Eduard TudoreanuUniversity of Arkansas at
  Little RockAR 722024, Little Rock, USAe-mail
  metudoreanu_at_ualr.edu
• Niklas ElmqvistDepartment of Computer
  ScienceChalmers University of Technology412 96
  Gothenburg, Swedene-mail elm_at_cs.chalmers.se

22
Outline

• Introduction
• Related work
• Preliminaries
• User tasks
• Selection of occlusion management techniques
• User study
• Results and discussion
• Conclusions

23
Related Work

• Many approaches for managing object density
• Multiple views additional 3D views to help
  disambiguate in areas of high object congestion
• Example WIM Stoakley et al. 1995, worldlets
  Elvins et al. 1997, birds eye views Fukatsu
  et al. 1998
• Space distortion object-space or view-space
  2D/3D distortion
• Example fisheye views Furnas 1986, hyperbolic
  tree browser Lamping and Rao 1996, non-linear
  projections Singh 2002, Singh and Balakrishnan
  2004
• Direct manipulation user-controlled invasive
  manipulation of objects
• Example 3D explosion probe Sonnet et al. 2004,
  BalloonProbe Elmqvist 2005, SDM Chuah et al.
  1995
• Transparency dynamic management of transparency
  to expose hidden content
• Example see-through surfaces Chittaro and
  Scagnetto 2001, view-dependent transparency
  Diepstraten et al. 2002, etc

24
Preliminaries

• We identify a number of representative user tasks
  for 3D virtual environments
• Discovery finding targets in a collection of
  objects
• Access retrieving information from targets
• Relation relating targets with each other and
  context
• We translate these to a number of concrete tasks
  to evaluate
• Count count the number of targets
• Pattern identify the pattern the targets form
• Relate find a third object spying on two
  targets (an object always in the neighborhood)

25
Preliminaries (2)

• Properties of occlusion management techniques
• Interactive and direct manipulation
• Designed for use by a user exploring the 3D world
• General-purpose (targets not previously known)
• Automatic or query-based methods are designed for
  a purpose
• One-phase interaction only
• Selection, filtering, and grouping disregarded
• Single-view methods
• Multiple views are difficult to manage in
  immersive VR
• Selection of representative techniques
• BalloonProbe, 3D fisheye views, standard 3D
  navigation

26
User Study (4)

• Subjects assigned to one of three tasks
• Each session lasted up to four hours
• Most subjects split the session into two or more
• Subjects were allowed to use a chair in the VE
• 1280 trials recorded in total