1. Image Space Based Visualization of Unsteady Flow on Surfaces
Robert S. Laramee1, Bruno Jobard2
Helwig Hauser1
1VRVis Research Center, Austria,
2University of Pau, France,

2. Overview goals previous work
image space based vis of unsteady flow on surfaces
results
conclusions, future work

3. Our Goals & Motivation application: vis of CFD simulation data
dense rep. of unsteady flow on surfaces
visualize flow on complex, adaptive resolution surfaces
user-interaction
fast
vis flow on dynamic meshes
no parameterization

4. Previous Work Lagrangian-Eulerian Advection by Jobard et al.
path integration (Lagrangian)
update of color pixels (Eulerian)
Image Based Flow Visualization by Van Wijk
advection of images

5. Method Overview Vector Field Projection Edge Detection
Compute Advection Mesh
Dynamic Case
Image Advection
Noise Blending
Static Case
Edge Blending
Image Overlay Application

6. Vector Field Encoding Velocity Image
Assign colors to the mesh vertices as a function of velocity
Velocity Image
Colored image used as the simplified (view dependent) 3D vector field
3D vectors are projected to image space transforming the computation from 3D to 2D
No more computation time spent on occluded polyons

7. Advection Mesh Computation and Boundary Treatment
Euler approximation of a pathline (like IBFV)
pk+1 = pk + vp(pk;t) dt
Advect Noise
Backward integration (like LEA)
Pk-1 = pk - vp(pk-1;t) dt

8. Edge Detection and Blending
Discontinuity Condition
|zk-1 - zk | > e |pk-1 - pk|

9. Edge Detection and Blending
can also be used to prevent background color(s) from “leaking in”
edge detection enabled

10. Noise Injection and Blending
Why noise injection and blending? –for full coveratge
both spatial and temporal characteristics: linearly interpolated sequence of random values
Temporal characteristics: a black and white pulsing function

11. Image Overlay Application
final stage
computed once for each dynamic case
applied last
User controlled opacity

12. Putting Pieces Together
Vector Field Projection
Edge Detection
Compute Advection Mesh
Image Advection
Noise Blending
Edge Blending
Image Overlay Application
Dynamic Case
Static Case

13. Texture Clipping Example is exaggerated for exposition
Texture clipped along edges to reduce artifacts

14. Results, Large, Complex Data Sets

15. Results, Large, Complex Data Sets

16. Results, Zooming

17. Results, Medical Simulation Data

18. Results, Performance Times
# mesh polygons
% image
space
Advection mesh res.
FPS static
FPS dynamic
10K 75
32^2 64 40
64^2 35
128^2 18
256^2 32 8
512^2
15-16
2.3
48K 74 13
10-11 6 15 2
221K 84 5 4
63-64
2.9
1.5
Not the same as in paper!
Nvidia 980XGL Quadro
2.8 GHz dual processor

19. Results, time-dependent mesh geometry and topology

20. Conclusions and Future Work
We presented an algorithm with the following properties:
dense representation of unsteady flow on surfaces
vis flow on large surfaces, independent of surface’s complexity and resolution
supports user-interaction
fast
vis flow on dynamic meshes
does not rely on parameterization
Future work
extension to unsteady, 3D flow
application of more specialized HW, e.g. programmable per-pixel operations

21. Acknowledgements Thanks to: Helmut Doleisch Tom Laramee Michael Mayer
Jürgen Schneider
Jarke van Wijk
Austrian governmental research program, Kplus (
AVL (
Many result animations available at: