Bullet Ray Vision
Lee A. ButlerUS Army Research Laboratory
Abe StephensUniversity of Utah
SCI Institute
Genesis• Contract to MAGI in 1966*
– Observation: projectiles passing through matter have similarity to photons passing through lenses.
• MAGI later developed Synthavision and did most of the rendering for TRON(1982).
Background• Photon transport was adapted to ballistic
penetration. The threat replaced the photon. The target replaced the lens.
• Ballistic penetration is like participating media. As the penetration occurs, there is interaction with the target media. Both the threat and the target are affected by the interaction.
• Computation is performed on the entire ray/object intersection, not just at the surface.– CSG was a natural geometric representation.
Original Design• Everything is expensive to compute, so
allow everything to be re-used– Ray-geometry intersection was computed
and saved for re-use. Typically, penetration equations and parameters were altered for each use.
– Assumes a single ray/threat relationship.
THOR• V50: Velocity at which 50% of fragments will
penetrate:V50(ft/sec) = 10c (h(in) Af(in
2))
Wf
(grains) sec
• Residual Velocity:
Vr(ft/sec) = V(ft/sec) – 10c (h Af)
Wf sec V
(ft/sec)
• Residual Weight:
Wr(grains) = Wf – 10c (h Af)
Wf sec V
Cultural Evolution• Programmer-user to Application-user.
• Data re-use to application re-run.
• Single ray/threat relationship to multiple rays/threat relationship.
• Ray tracing slow to fast
One Step Forward
Ray Depth Top Right Front A35Full 9.22 10.52 6.27 12.36Adaptive 7.30 10.54 3.15 7.60
View Computation Time (seconds)
• Interleave ballistic penetration calculations with ray/geometry intersection using classic BRL-CAD ray-tracer on CSG geometry.
• Nice performance improvement, but still slow.
Two Steps Forward• Bullet Vision:
Since we’re ray tracing,
render the results of the
computation too.
Free visualization!
Three Steps Forward• Implement computation as a shader in Manta
packet-based ray tracer.
– Need to collect in/out pairs before “shading.”
• What was a batch application is now
interactive. 3.9 fps on Intel Core 2 2.66Ghz
with 4 cores.
• Surprise: BVH Traversal is the major
bottleneck in the system. “Shader” with 14
exponentiation operations is distinct second.
Future Work• Current work is a brute-force
implementation of the penetration equations. 14 exponential operations.
• There is ample opportunity to optimize the computation of the equations.
• This is a equation fit to measured data. Alternative equation fits that are more computationally friendly are possible.
Future Work• BVH acceleration structure may not be
optimal for applications where “transparent” geometry is the norm. Some further investigations on acceleration structures for such applications is needed.
• Current frustum acceleration techniques may not be optimal for transparency and deferred shading algorithms.