Employing Dynamic Transparency for 3D Occlusion Management: Design Issues and Evaluation

981 views
850 views

Published on

Slides from INTERACT 2007 on dynamic transparency for 3D virtual environments.

Published in: Technology
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
981
On SlideShare
0
From Embeds
0
Number of Embeds
45
Actions
Shares
0
Downloads
8
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide
  • Thanks for the introduction and good afternoon. As stated, my name is Niklas Elmqvist and I am a postdoc at INRA in Paris, France. This work was done in collaboration with colleagues at Chalmers University of Technology in Goteborg, Sweden: Ulf Assarsson and Philippas Tsigas (my dissertation advisor). Looking at the title of this talk, it is perhaps not obvious what I am going to be discussing today.
  • The following few images might shed some light on the problem I am attacking. The question is what do these pictures have in common? First, a cocktail party. Second, a 3D environment in a popular computer game. Third, a volume visualization of stress measurements in a tunnel under construction. And fourth, a molecule visualization in 3D. Well, there are a lot of things that these pictures may have in common, but the answer I am looking for here is occlusion. Occlusion, or the fact that objects in these highly detailed 3D environments, be they real or virtual, end up hiding each other, or occluding each other.
  • Why is occlusion an important issue? Well, if these 3D pictures that we just saw represent visualizations where the objects are information-carrying entities, we may have a problem. IIRVEs is a concept dealing with the combination of information visualization and 3D environments. There is a lot of potential in this combination---for one thing, we have a new dimension to use for our visualizations---but there are a number of problems as well. Mostly, beyond the complexity of navigating 3D environments, which can be tricky, these come down to the issues of visibility and legibility of objects. This in turn affects the performance of users wanting to discover objects in an environment, accessing information encoded in them, and relating them to other objects. Occlusion is one of the major causes for these problems. Where in 2D visualizations, you can avoid occlusion by just avoiding object intersection in the 2D plane, it is not as simple in 3D worlds: even if objects do not intersect, they may end up hiding each other. A small object close to the 3D camera may occlude a whole scene. This is the case for the cocktail party picture I showed earlier: entering a crowded party filled with mingling people, you may have hard time finding your friends and you may have to do an exhaustive search to locate them.
  • Now, given that occlusion is an issue, we would like to address it. The inspiration for this work is actually quite simple: what if we could give our users Superman-like abilities so that they can see through objects and discover what they are looking for? Obviously, this is hard to do in the real world, like the cocktail party in the beginning of the talk, but it is actually quite possible in the computer world and 3D environments we are dealing with. So basically, our idea is to give our users superhuman-vision that allows them to see through walls, see things far away, and see things that are too small to see with the naked eye.
  • Now, allow me to make a short sidenote about Superman and his X-ray vision. Here are a few pictures showing this in action: as you can see, Superman has the ability to see through any kind of material (except lead) to discover things hidden behind. For instance, he can discover a pistol in a bad guy’s pocket, or a burglar breaking into his apartment. All in all, a nice ability to have.
  • So the basic idea is simple: we want to provide our users with this kind of X-ray vision. The benefit would be to decrease the impact of occlusion on the visibility and legibility of objects in the 3D world. It would allow our Superman-users to pinpoint important targets in the 3D worlds despite occluding distractors. As we have already seen, this is a major problem for 3D visualization, but perhaps we can address it this way?
  • If we adapt the idea a little bit more to computer graphics, the basic idea is simple: we dynamically adjust the transparency of intervening surfaces so that targets hidden behind them are made visible. Obviously, this idea is not 100% novel, and some other researchers have had similar ideas. Some existing such techniques include perspective cutouts by Coffin and Höllerer (presented at the IEEE 3DUI symposium in 2006), where CSG techniques are used to cut out pieces of this wall to show the hidden teapot. Interactive break-away views, presented by Diepstraten and others at Eurographics in 2003, exposes occluded objects, such as the ball and the teapot in the pipe, by cutting out a region of the surface in the image-space using programmable graphics hardware. The importance-driven volume rendering approach by Viola, presented at IEEE Vis in 2004, assigns importance to all voxels in the volume model and then renders the scene given these importance values, in this case exposing the internal organs of the poor gecko. The common denominator of all of these works, and other work as well, is that there is no user evaluation on how dynamic transparency affects users in solving visual perception tasks in these 3D environments. Yes, obviously they will help in finding the targets, but it is also important that dynamic transparency both decreases the depth cues in the scene as well as increases the visual complexity. What impact will this have on user performance?
  • To be able to answer this question, we first formulate a general model of dynamic transparency. This is summarized in the following four rules. The first one captures the core of X-ray vision, and simply states that all targets should be visible from all viewpoints, regardless if they are hidden by distractors or not. The second rule stipulates that targets are made visible using a cutout area where the transparency is changed from full opacity to semi-transparency---an alpha value that is a little bit bigger than 0%. We have already seen this cutout area in effect in the earlier Superman cartoons, where Superman would have a circular cutout area to show what is behind a particular surface. Also, just like lead for Superman, we think it is useful to be able to have the notion of an impenetrable surface that will not be made transparent. For instance, it does not make sense to make the border surface around a visualization visible, so we want to be able to mark them as impenetrable. Finally, the fourth rule states that targets are allowed to self-occlude. In other words, if the handle of a teapot is hidden behind the teapot itself, we don’t make the teapot semi-transparent to show it. Otherwise, the visual output may become very complex and hard to understand.
  • In this work, we define a technique for dynamic transparency based on the observation that the image space is perfect for detecting occlusion: whenever we find ourselves overdrawing a pixel belonging to a target, we know that occlusion is occurring. In our technique, we employ fragment and vertex shaders to achieve a Superman-like effect exposing the targets. This is done by rendering the targets to separate off-screen buffers and alpha-blending them back onto the frame buffer to get the desired effect.
  • Here is a screenshot showing the technique in action. On the left we see an image of a 3D model of a jeep. If we mark the engine inside the jeep as a target and turn on image-space dynamic transparency, we get the following result. Regardless of how the user moves, the engine will now be visible through the hood of the car, allowing the user to see its relationship to the rest of the vehicle.
  • In order to answer our questions from the beginning of this talk of whether or not dynamic transparency will have an adverse impact on user productivity, we conducted a user study. Our hypothesis was that users would perform their visual perception tasks more efficiently than without. In other words, we believed that the loss of depth cues and the increased visual complexity that dynamic transparency causes would not have a significant impact on user efficiency. Given this objective, we performed a comparison for standard 3D camera navigation with dynamic transparency available or not. We recruited 16 participants to perform the study, 13 of which were male and 3 which were female. The single factor we investigated was whether dynamic transparency was on or off.
  • To try to increase the generality of our findings a little, we performed the study in two different kinds of 3D worlds: first, an abstract 3D world consisting of 3D primitives such as the picture to the right shows. The motivation for this was abstract infovis applications. The two tasks we had our subjects perform was first to count the number of targets (red cones) in the environment, and then to identify the global pattern they formed. As you can see, with dyntrans active, all of the red objects were tagged as targets, and were thus visible despite occluding distractors. For the second world, we targeted a more realistic-looking environment: a virtual walkthrough application. This was simple a one-floor building consisting of rooms, walls, and doors and a number of objects scattered around the rooms. With dynamic transparency active, the objects were tagged as targets and were thus visible through the walls, just like how Superman would see the world. In the third task, participants was asked to find a unique target in the world and then mark its location on the map, and in the fourth task, they were asked to count the number of targets of a specific kind.
  • An analysis of the data from the user study gives the following results: For completion time, we found that participants were significantly faster for all tasks using dynamic transparency than without. The bar chart summarizes these findings. The differences were significant down to p < 0.05. For errors, we got a slightly different result. For task 1, 2 and 4, we collected the ratio of errors per total number of targets. Only for T1 were participants significantly more accurate using dynamic transparency, not for the other ones. The bar chart shows the errors.
  • Finally, with task 3 we wanted to study whether the fact that users with dynamic transparency would be able to solve the find task faster would instead cause them to lose accuracy. In other words, if they are able to see the target through the walls of the building, maybe they would instead be worse at placing the target on the map than a user force to navigate the rooms in turn. However, as the chart shows, we found no significant difference in the distance users placed the target from its actual position on the map. This supports us in believing that users do not lose too much depth information using the technique. Having said that, it is still important to recognize that occlusion IS an important depth cue, and that reducing it may cause loss of depth information. In particular, we need to avoid the effect of reverse occlusion, where distant objects hide nearby objects. In our solution, we use the gradient outline of the cutout shape and the fact that the cutout retains an alpha threshold to combat this. In other words, if you are looking at a car through a brick wall, you will still see a faint brick wall pattern on the car. One interesting observation that we found while during the study was that users had a tendency to respect the 3D world more if dynamic transparency was not active. We had no collision detection in the test application, and we often found that participants would just fly through walls if dynamic transparency was active, whereas they would use the doors if it was not active. This is important to keep in mind---we lose some visual realism with dynamic transparency.
  • So, in conclusion, in this talk I have put forward the notion of using Superhero X-ray vision to combat the effects of occlusion on user productivity in 3D visualization. Our model for dynamic transparency supports this mechanism by making sure that targets are always visible regardless of occluding distractors. I have also reported on a user study where we found that users were indeed faster and generally as accurate or more accurate using dynamic transparency in comparison to not having it. Still, as I mentioned in the previous slide, occlusion does provide important depth cues, and reducing them may hamper the user’s depth awareness. This is an important caveat to keep in mind when using the technique.
  • Thank you for your attention and I will be happy to take your questions.
  • Employing Dynamic Transparency for 3D Occlusion Management: Design Issues and Evaluation

    1. 1. Employing Dynamic Transparency for 3D Occlusion Management: Design Issues and Evaluation Niklas Elmqvist < [email_address] > Ulf Assarsson < [email_address] > Philippas Tsigas < [email_address] > INRIA Saclay, France Chalmers University of Technology Gothenburg, Sweden
    2. 2. The Least Common Denominator… Employing Dynamic Transparency 3D Occlusion Management: Design Issues and Evaluation Occlusion!
    3. 3. Problem: 3D Visualization <ul><li>Information-Rich Virtual Environments (IRVEs) </li></ul><ul><ul><li>Information visualization in 3D </li></ul></ul><ul><ul><li>[Bowman et al. 2003] </li></ul></ul><ul><li>IIRVE has a lot of potential but is tricky </li></ul><ul><ul><li>Visibility and legibility of objects </li></ul></ul><ul><ul><ul><li>Discover objects </li></ul></ul></ul><ul><ul><ul><li>Access information encoded in objects </li></ul></ul></ul><ul><ul><ul><li>Spatially relate objects </li></ul></ul></ul><ul><li>Occlusion is one of the main causes </li></ul><ul><li>Particularly problematic for 3D visualizations </li></ul><ul><ul><li>Easier in 2D, but still… </li></ul></ul><ul><ul><li>” Cocktail party” effect </li></ul></ul>Employing Dynamic Transparency 3D Occlusion Management: Design Issues and Evaluation
    4. 4. Inspiration <ul><li>What if we could endow all human users with Superman-like powers of observation? </li></ul><ul><ul><li>Difficult in the real world </li></ul></ul><ul><ul><li>Possible in the computer world </li></ul></ul><ul><li>Idea : Give the users super-human vision </li></ul><ul><ul><li>See through walls </li></ul></ul><ul><ul><li>See things far away </li></ul></ul><ul><ul><li>See things too small to see with the naked eye </li></ul></ul>Employing Dynamic Transparency 3D Occlusion Management: Design Issues and Evaluation
    5. 5. Example: Superman’s X-Ray Vision Employing Dynamic Transparency 3D Occlusion Management: Design Issues and Evaluation &quot;Where we come from everyone has see-through vision, extra-strength and extra-speed!“ [S No. 65/3: &quot;Three Supermen from Krypton!“]
    6. 6. Benefits <ul><li>Let us provide our users with X-ray vision! </li></ul><ul><li>X-ray vision has a very important benefit: </li></ul><ul><ul><li>Avoids the previous problems with visibility and legibility in 3D environments </li></ul></ul><ul><ul><li>Can easily pinpoint important targets despite occluding distractors </li></ul></ul><ul><li>Main stumbling block of 3D information visualization </li></ul><ul><ul><li>Caused by the nature of the human vision system </li></ul></ul><ul><ul><ul><li>(But not the superhuman vision system...?) </li></ul></ul></ul>Employing Dynamic Transparency 3D Occlusion Management: Design Issues and Evaluation
    7. 7. Dynamic Transparency <ul><li>Idea : Adjust transparency of surfaces to make targets visible through occluding distractors </li></ul><ul><li>Existing techniques for dynamic transparency </li></ul><ul><ul><li>Perspective cutouts [Coffin and Höllerer 2006] </li></ul></ul><ul><ul><li>Interactive break-away [Diepstraten et al. 2003] </li></ul></ul><ul><ul><li>IDVR [Viola et al. 2004] </li></ul></ul><ul><li>No user evaluations have been performed </li></ul>Employing Dynamic Transparency 3D Occlusion Management: Design Issues and Evaluation
    8. 8. Dynamic Transparency Model <ul><li>We define our model for dynamic transparency as a set of rules: </li></ul><ul><ul><li>R1 : All important objects (targets) in a scene should be visible from any given viewpoint </li></ul></ul><ul><ul><li>R2 : Targets are made visible by changing the transparency level of occluding surfaces from opaque (  = 100%) to transparent (  =  t > 0%) within a cutout area enclosing the object </li></ul></ul><ul><ul><li>R3 : Some surfaces are impenetrable and will never be made transparent (cf lead for Superman) </li></ul></ul><ul><ul><li>R4 : Targets are allowed to self-occlude themselves </li></ul></ul><ul><li>Cutout area : convex hull (circle) or outline with a gradient transparency border </li></ul>Employing Dynamic Transparency 3D Occlusion Management: Design Issues and Evaluation
    9. 9. Image-Space Dynamic Transparency <ul><li>Observation : The image space is perfect for detecting instances of occluded targets and dynamically adjusting transparency to allow the user to ”see through” surfaces </li></ul><ul><ul><li>Can employ fragment and vertex shader capabilities of modern programmable graphics hardware </li></ul></ul><ul><ul><li>Achieve Superman-like ”cutaway effect” of surfaces to retain depth cues and spatial information </li></ul></ul><ul><li>Our algorithm renders targets into an offscreen buffer and alpha blends on frame buffer to achieve Superman-like X-ray vision </li></ul>Employing Dynamic Transparency 3D Occlusion Management: Design Issues and Evaluation
    10. 10. Screenshots Employing Dynamic Transparency 3D Occlusion Management: Design Issues and Evaluation
    11. 11. User Study <ul><li>Hypothesis : Users perform visual perception tasks better with dynamic transparency </li></ul><ul><ul><li>(Loss of depth cues and increased visual complexity will not be a major factor) </li></ul></ul><ul><li>Comparison : standard 3D camera navigation </li></ul><ul><li>Subjects : 16 paid participants (13 male, 3 female) </li></ul><ul><li>Factors : dyntrans </li></ul><ul><ul><li>Dynamic transparency on or off </li></ul></ul><ul><li>Repeated-measures within-subject design </li></ul>Employing Dynamic Transparency 3D Occlusion Management: Design Issues and Evaluation
    12. 12. Tasks and Worlds <ul><li>Abstract 3D World: </li></ul><ul><ul><li>Count number of targets </li></ul></ul><ul><ul><li>Identify the pattern formed by targets </li></ul></ul><ul><li>Virtual Walkthrough : </li></ul><ul><ul><li>Find unique target </li></ul></ul><ul><ul><li>Count number of targets </li></ul></ul>Employing Dynamic Transparency 3D Occlusion Management: Design Issues and Evaluation
    13. 13. Results <ul><li>Completion time </li></ul><ul><li>Averages for all tasks: </li></ul><ul><ul><li>Standard: 65 seconds </li></ul></ul><ul><ul><li>Dyntrans: 29 seconds </li></ul></ul><ul><ul><li>Significant (p < 0.05) </li></ul></ul><ul><li>Errors </li></ul><ul><li>Task 1, 2, 4: errors per total number of targets </li></ul><ul><li>T1 significant, others not </li></ul>Employing Dynamic Transparency 3D Occlusion Management: Design Issues and Evaluation
    14. 14. Discussion <ul><li>Task 3 shows the accuracy of marking an object on a 2D map </li></ul><ul><ul><li>Dyntrans has no adverse effect on depth cues </li></ul></ul><ul><li>Occlusion is still an important depth cue </li></ul><ul><ul><li>Avoid “reverse occlusion”! </li></ul></ul><ul><ul><li>Use cutout shape + other cues </li></ul></ul><ul><li>Observation : Users respect world more with no dyntrans </li></ul>Employing Dynamic Transparency 3D Occlusion Management: Design Issues and Evaluation
    15. 15. Conclusions <ul><li>Superhero X-ray vision has an important benefit </li></ul><ul><ul><li>Avoids visibility and legibility problems by allowing for occluding surfaces to be made (semi-)transparent </li></ul></ul><ul><li>Our model for dynamic transparency supports this mechanism in visualization applications </li></ul><ul><ul><li>Targets are always visible through semi-transparent cutouts in occluding distractors </li></ul></ul><ul><li>Results from our user study: </li></ul><ul><ul><li>Dynamic transparency allows for solving visual perception tasks faster and with generally better or equal accuracy to standard 3D navigation </li></ul></ul><ul><li>Depth cues is an issue… </li></ul>Employing Dynamic Transparency 3D Occlusion Management: Design Issues and Evaluation
    16. 16. Questions? <ul><li>Contact information: Niklas Elmqvist INRIA, Saclay, France E-mail: [email_address] Web: http://www.lri.fr/~elm/ Phone: +33 1 69 15 61 97 Fax: +33 1 69 15 65 86 </li></ul>Employing Dynamic Transparency 3D Occlusion Management: Design Issues and Evaluation
    17. 17. On Superhero X-Ray Vision <ul><li>Sources : Supermanica ( supermanica.info ) and the Superman Encyclopaedia ( theages.superman.ws/Encyclopaedia/ ) </li></ul><ul><li>Major components: </li></ul><ul><ul><li>X-ray vision : see through all substances and materials except lead </li></ul></ul><ul><ul><li>Telescopic vision : see (very) distant objects </li></ul></ul><ul><ul><li>Supervision : combination of x-ray and telescopic vision </li></ul></ul><ul><ul><li>Microscopic vision : see on a microscopic scale </li></ul></ul>Employing Dynamic Transparency 3D Occlusion Management: Design Issues and Evaluation “ Today's Superman possesses a wide range of optical super-powers, including X-ray vision , which enables him to see through all substances except lead; telescopic vision , which enables him to focus on objects millions of miles away; super-vision , a combination of X-ray vision and telescopic vision, which enables him to perform such optical feats as peering through the wall of a house thousands of miles away; micro-scopic vision , which enables him to examine the tiniest atomic particles…”

    ×