Upload
matthias-trapp
View
1.352
Download
1
Embed Size (px)
Citation preview
Enhancing Interactive Non-Planar Projections
of 3D Geovirtual Environments with Stereoscopic Imaging
Matthias Trapp, Haik Lorenz, Markus Jobst, Jürgen DöllnerHasso-Plattner-Institute at the University of Potsdam
True-3D in Cartography1st International Conference on 3D Maps
August 24 - 28, 2009 Dresden, Germany
1
motivation geo-media technology
provides interactivity, immersion facilitates the communication of 3D geo spatial data
applications to cartography: increase immersion into 3D geovirtual environments support for depth-cues
planar stereoscopy: well understood – rendering: straight forward supported by graphics hardware / driver
non-planar stereoscopy: provides high field-of-view and image resolution 2
problem: non-planar projection surfaces
rendering of digital 3D city and landscape models: high amount of geometry and texture data real-time constraints (> 20 frames per second)
current generation of graphics hardware (GPU) no native support for non-planar projection surfaces requires specific rendering techniques classified into image, geometry, and ray-based
approaches
hardware-accelerated stereoscopic imaging: available stereo hardware modifies vertex pipeline stage cannot be used for rendering non-planar
stereoscopy3
framework - conceptual overview
1..N Stereo Mates
Stereo Rendering Component
SceneGeometry + Textures 1..N Virtual Cameras
Stereo Mate Generation
Chroma Depth Active
Output Device (Screen, Projector, Printer)
Passive ...
Planar Projection Image-basedNon-Planar Projection
Geometry-basedNon-Planar Projection
4
review: image-based approach (IBA)
basic concept: dynamic cube map + screen-aligned quad image warping based on normal vectors:
3-phase rendering process:1. create/update dynamic cubemap2. setup projection shader3. render screen-aligned quad -β/2
-α/2
Fst = (s,t)β/2
0
α/2
φ
Viewport
5
adapting IBA for stereoscopy basic idea for image-based non-planar projections:
create cubemaps for each virtual camera derive non-planar projection for each cube-map
examplary workflow for two stereo mates:
Layered Rendering Projection Function δP + Layer Sampling
Polygonal Scene Texture Layers Non-Planar Projection Stereo PairsLeft
Right
6
review: geometry-based approach (GBA)
projection computed on a per-vertex basis ensure sufficient on-screen vertex density dynamic mesh refinement required
primitive index (e.g.
0,0,1,2,2,2,…)
step 3: indexed renderingrender each indexed primitive
into projection pieces
projection & clip matrices framebuffer
step 2: primitive replication create an index containing the respective number
of replications per input triangle
scene triangles
per-triangle replication
count
step 1: replication determination calculate and write one replication count per input
triangle
COP COP
Non-Planar Projection Surface Non-Planar Projection SurfaceApproximation (coarse)
7
adapting GBA for stereoscopy straight forward approach:
setup piece-wise projection for each virtual camera render into different color-buffers additional post-processing step: layer compositing
example for stereo image pairs:
step 3: indexed renderingrender each indexed primitive into two projection
pieces (for left and right eye)
scene triangles
per-triangle replication
count
step 1: replication determination calculate and write one replication count per input
triangle
step 2: primitive replication create an index containing the respective number
of replications per input triangle
left eye framebuffer
projection & clip matrices
primitive index (e.g.
0,0,1,2,2,2,…)
right eye framebuffer
step 4: stereo compositingjoin both framebuffers into a single output image
framebuffer
8
rendering active & passive stereo active stereo:
using quad-buffering usually encapsulated by graphics driver
passive stereo: anaglyph: color-buffer compositing polarized: render to framebuffer chromo-depth stereo: apply directly during rendering
Left Right
9
rendering chromo-stereoscopy
nearv
rm = ƒ
rm (vclam
p )
vdt = ƒdt(V)
vclamp = ƒclamp(vdt) far
RG
B
CM
Y
RW
B
Cout =
ƒsam
ple (vrm )
cs ce
no need for generating stereo image pair color as a function of depth
11
applying chromo-stereoscopy GBA: straight forward application to fragment‘s
depth IBA: needs depth correction
13
chroma-stereoscopy issues common problems for IBA and GBA:
distribution of color can decrease stereo effect perception: facade information (texture) is altered interaction: focal plane must be adapted
Near Focal Plane Far Focal Plane
Equally Distributed
15
binary comparision GBA vs. IBAComparison Criteria GBA IBA
Stereo Functionality Image Quality Rendering Performance Memory Footprint Implementation Complexity Overall Rating
16
conclusions & future work conclusions:
interactive stereoscopic rendering for non-planar projections
increases immersion, thus psychological depth cues performance limited by geometric complexity of the
scene GBA outperforms IBA but IBA much easier to
implement/use
open problem: omni-directional stereo without image artifacts
future work: auto stereoscopy for non-planar projections surfaces eye tracking to adjust user‘s focal plane 17
Thank you for your attention! Questions?
Contact Matthias [email protected]
Haik [email protected]
Markus [email protected]
Jürgen Dö[email protected] Workgroup 3D Geoinformationwww.3dgi.de/
Computergraphics System Groupwww.hpi.uni-potsdam.de/doellner/
18