Combining Stereo and Visual Hull Information for On-line Reconstruction and Rendering of Dynamic Scenes
Dec 11th, 2002 Ming Ming Li, Li, Hartmut Hartmut Schirmacher Schirmacher Marcus -Peter Seidel Marcus Magnor Magnor,, Hans Hans-Peter Seidel MMSP’02
Ming Li
Introduction
INFORMATIK
The general goal •• Acquiring -world scenes Acquiring dynamic dynamic 3D 3D geometry geometry of of real real-world scenes and -time and render render itit in in real real-time
Previous systems •• Visual Visual hull hull –– Image -Based VH Image-Based VH & & Polyhedral Polyhedral VH VH (MIT) (MIT) •• Stereo Stereo –– 3D -presence ((UPenn.) UPenn.) 3D Dome Dome (CMU), (CMU), Tele Tele-presence
Our work •• Combine Combine two two methods methods at at interactive interactive frame frame rate rate MMSP’02
Polyhedral visual hull reconstruction •• Project Project silhouette silhouette polygons polygons back back to to 3D 3D space space •• CSG CSG intersection intersection MMSP’02
Ming Li
Depth from stereo
INFORMATIK
Rectification •• Warps Warps the the stereo stereo images images to to align align them them along along scanlines scanlines
Matching
Dis pari ty Se arch Rang e
Depth recovery Depth =
MMSP’02
Baseline ∗ FocalLengt h Disparity Ming Li
Combination of Stereo & VH
INFORMATIK
Stereo
Visual Hull
Essence
rays radiating from inner surfaces of objects
rays radiating from tangent surfaces of objects
Pros
accurate geometry
robust
Cons
noisy result
approximate geometry
Combine them! •• Use Use visual visual hull hull to to stabilize stabilize stereo stereo recovery recovery •• Use Use stereo stereo algorithm algorithm to to recover recover concave concave geometry geometry details details MMSP’02
Ming Li
Combination of Stereo & VH
INFORMATIK
Basic Idea •• Exploit Exploit visual visual hull hull to to restrict restrict disparity disparity search search range range in in stereo stereo computation computation –– Disparity Disparity range: range: two two disparity disparity maps, maps, upper upper and and lower lower bound bound
Disparity range map from VH •• Use Use OpenGL OpenGL to to render render VH VH into into depth depth buffer buffer •• Convert Convert the the depth depth buffer buffer to to disparity disparity map map MMSP’02
Ming Li
System overview
INFORMATIK
Acquisition Acquisition Setup Setup •• 66 cameras cameras are are grouped grouped pairwise pairwise,, pointing pointing to to aa common common region region •• Pre -calibrated using Pre-calibrated using Tsai’s Tsai’s algorithm algorithm •• Synchronized Synchronized via via external external trigger trigger
MMSP’02
Ming Li
Processing stages
INFORMATIK
Initialization •• Compute Compute aa reference reference background background image image for for segmentation segmentation of of videos videos from from all all views views
Running system •• Processing Processing cycle cycle for for each each frame frame
MMSP’02
Ming Li
Improved reconstruction depth depth map map recovered recovered without without using using visual visual hull hull
MMSP’02
INFORMATIK
Improved Improved depth depth map map with with the the help help of of visual visual hull hull
Ming Li
Rendering
INFORMATIK M
3D forward warping •• Choose Choose one one closest closest view view
p1 ( X1, y1) p2 (X2, y2)
R ef erence vi ewpoi nt
Rendering snapshots
MMSP’02
D esir ed vi ew point
Ming Li
Results
INFORMATIK
System Configuration •• Clients Clients — — Three Three AMD AMD Athlon Athlon 1.1GHz 1.1GHz machines machines •• Server -processor machine Server — — Intel Intel 1.7GHz 1.7GHz dual dual-processor machine •• Input Input image image resolution resolution — — 320x240 320x240 pixels pixels –– About About 18,000 18,000 silhouette silhouette pixels pixels for for 3D 3D warping warping
Performance — Interactive frame rate •• Up Up to to 44 fps fps for for reconstruction reconstruction •• 10 10 fps fps for for rendering rendering MMSP’02
Ming Li
Video show
MMSP’02
INFORMATIK
Ming Li
Conclusion & Future work
INFORMATIK
Conclusions •• Better Better results results using using combined combined reconstruction reconstruction method method •• Reconstruction Reconstruction and and rendering rendering of of dynamic dynamic scenes scenes are are performed performed at at interactive interactive frame frame rates rates
Future work •• Use Use visual visual hull hull reconstruction reconstruction as as aa base base method method and and improve improve the the 3D 3D model model of of visual visual hulls hulls using using stereo stereo information information
Disparity range: two disparity maps, upper and lower bound. Disparity range map from VH. ⢠Use OpenGL to render VH into depth buffer. ⢠Convert the depth ...
monocular SLAM system which combines the benefits of these two techniques. ... that end we recast the usual world-centric EKF implementation of visual SLAM ...
retrieve images using semantic keywords and auto-annotate images based on ... can learn (about) natural language from training data such as document ...
Another advantage is that stereo is a passive ... Computing Surveys, VoL 14, No. 4, December ...... conditions, cloud cover present in one im- age and not in the ...
For correspondence: S. T. Barnard, Artificial Intelligence Center, SRI ... notice is given that copying is by permission of the Association for Computing Machinery. To ... 3.2 Control Data CorporatJon ...... conditions, cloud cover present in one im-
an unstructured random effect with one level for each observed mating. We illustrate the methods with the following simulated data. Note that in this example the ...
Jul 19, 2010 - Self-Training for Multilingual Text Categorization. Massih-Reza .... text classification. Section 4 describes the boosting-based algorithm we developed to obtain the language-specific clas- sifiers. In Section 5, we present experimenta
tures - Funge's cognitive architecture and the recent SAC concept. Addi- tionally it puts emphasis on strong design and provides easy co-operation of different ...
Mobile Multi-Sensor Systems Research Group. Department of ... ity and ease of implementation; however, a more fundamental fusion of the GPS data into the.
Architectural simulators like SimpleScalar [1] (and its derivatives), SMTSim [17] or Simics [13] employ a very simple technique for functional simulation. They normally employ interpreted techniques to fetch, decode and execute the instructions of th
Feb 4, 2009 - Keywords-Cloud computing; virtualization; mapreduce; bioinformatics. .... National Center for Biotechnology Information. The parallelization ...
the confluence of its east and west branches at. Hancock, New ... our team. The assigned usage levels for the access points remained fixed within each of the seven time blocks. ... during each calendar week; sampling of low-usage sites was ...
Digital Compass- OS5000-USD. ⢠Cameras- Allied Vision Prosilica GC750 (x3) two front-facing, one down-facing. ⢠Batteries- Thunder Power RC G4 Pro Power 45C. ⢠Hydrophones- Aquarian H1a. ⢠Depth Sensor- SSI part P51-15-S-B-I36-4.5V-R. Stingra
surveys produced similar estimates of boat angler effort and little evidence of bias, but shore anglers were undercounted in the aerial ... Published online August 28, 2006. 727 ...... From a statistical point of view, the price to pay for employing
We extend photometric stereo to make it work with in- ternet images, which are typically associated with differ- ent viewpoints and significant noise. For popular tourism sites, thousands of images can be obtained from internet search engines. With t
a Department of Linguistics and Modern English Language, Lancaster University, Lancaster LA1 4YT, United Kingdom ... Available online 19 March 2005. Abstract ...... MA. Maynard, D., Ananiadou, S., 2000. Trucks: a model for automatic multiword term re