INFORMATIK

Online Accelerated Rendering of Visual Hulls in Real Scenes February 6th, 2003

Ming -Peter Seidel Ming Li, Li, Marcus Marcus Magnor Magnor,, Hans Hans-Peter Seidel Max Max-Planck Planck--Institute for Max-Planck-Institute for Computer Computer Science Science Saarbruecken Saarbruecken, Saarbruecken,, Germany Germany

WSCG‘03

Ming Li

Talk Outline

INFORMATIK

Motivation System overview Visual hull reconstruction Rendering improvements Performance & Results Conclusions & Future work WSCG‘03

Ming Li

Introduction

INFORMATIK

What do we want? •• Acquire -world object Acquire dynamic dynamic 3D 3D geometry geometry of of real real-world object on -line and -time on-line and render render itit in in its its environment environment in in real real-time –– 3D 3D interactive interactive TV, TV, computer computer games, games, tele tele-communication communication and and sports sports analysis analysis etc. etc.

Visual hull •• An An approximate approximate shell shell that that envelopes envelopes the the true true geometry geometry of of the the object object •• Suitable Suitable for for real real time time reconstruction reconstruction and and rendering rendering

WSCG‘03

Ming Li

Motivation

INFORMATIK

Our work •• Develop Develop improved improved rendering rendering algorithms algorithms accelerated accelerated by by graphics graphics hardware hardware •• Background Background rendering rendering –– Noticeable Noticeable boundary boundary seams seams between between multiple multiple views views •• Visual Visual hull hull rendering rendering –– Texture Texture will will project project through through

WSCG‘03

Ming Li

System Overview •• ••

INFORMATIK

Clients Clients — — Three Three AMD AMD Athlon Athlon 1.1GHz 1.1GHz machines machines Server -processor machine Server — — Intel Intel P4 P4 1.7GHz 1.7GHz dual dual-processor machine with with GeForce GeForce 33 graphics graphics card card

•• Six Six cameras cameras are are arranged arranged roughly roughly along along aa circle circle •• Pre -calibrated using Pre-calibrated using Tsai’s Tsai’s algorithm algorithm •• Sync Sync via via external external trigger trigger •• Image Image resolution resolution — — 320x240 320x240 pixels pixels WSCG‘03

Ming Li

Visual Hull Reconstruction

INFORMATIK

Silhouette extraction •• Static Static background background acquired acquired beforehand beforehand •• Background Background differencing differencing &Polygon &Polygon approximation approximation

Polyhedral visual hull reconstruction •• Project Project silhouette silhouette polygons polygons back back to to 3D 3D space space •• 3D 3D intersection intersection WSCG‘03

Ming Li

Background Blending

INFORMATIK

Basic approach •• Project Project multiple multiple views views onto onto aa background background geometry geometry proxy proxy N 1 •• Target Target blending blending function: function: C = N * ∑ C k

Problem

k =1

•• Need Need per-pixel per-pixel accumulation accumulation and and normalization normalization

Solution •• Frame Frame buffer buffer blending blending for for accumulation accumulation •• Alpha Alpha channel channel for for pre-normalization pre-normalization WSCG‘03

Ming Li

Background Blending Details

INFORMATIK

•• Multi-pass Multi-pass rendering rendering and and accumulate accumulate blended blended results results in in frame frame buffer buffer •• Initial Initial frame frame buffer buffer (0, (0, 0, 0, 0, 0, 1), 1), SRC_ALPHA=0.5 SRC_ALPHA=0.5 •• Color Color channel channel blending blending C = C s * DST _ALPHA + C d * (1 − DST _ALPHA )

–– This This formula formula computes computes average average color color IfIf DST_ALPHA’s DST_ALPHA’s form form reciprocal reciprocal number number series series •• Alpha Alpha channel channel blending blending A = As * 0 .12 + Ad * SRC _ ALPHA

–– Approximate Approximate reciprocal reciprocal number number series series WSCG‘03

Ming Li

Background Blending Results

Without Blending

WSCG‘03

INFORMATIK

With Blending

Ming Li

Projective Texture Mapping & Shadow Mapping

INFORMATIK

Normal projective texture mapping •• Occluded Occluded parts parts also also get get textured textured

Shadow mapping •• Render - of-view Render from from the the light’s light’s point point-of-view to to generate generate depth depth map map •• Render -of-view Render from from the the eye’s eye’s point point-of-view –– Use Use depth depth map map as as aa projective projective texture texture •• Produce Produce texel texel 11 for for lit lit parts parts ,, 00 for for shadowed shadowed parts parts WSCG‘03

light Zd in depth map

eye

Z position relative to light Ming Li

Opaque Projective Texture Mapping

INFORMATIK

Opaque projective texture mapping •• Multiple Multiple textures textures –– Tex0: Tex0: Color Color texel texel,, Tex1: Tex1: Shadow Shadow texel texel –– Final Final fragment: fragment: Tex0*Tex1 Tex0*Tex1 •• Enable Enable alpha alpha testing testing to to remove remove occluded occluded parts parts

Problem

Shadow leak

•• Shadow Shadow leak leak due due to to insufficient insufficient sampling sampling of of depth depth map map WSCG‘03

Ming Li

Extended Opaque Projective Texture Mapping

INFORMATIK

Two options •• Increase Increase depth depth map map resolution resolution –– Cost Cost more more rendering rendering time time –– Consume Consume more more texture texture memory memory •• Render Render aa fatter fatter version version of of depth depth map map –– 22-pass -pass depth depth map map rendering rendering 8 811stst Pass: Pass: Line Line mode mode using using thick thick line line width width nd Pass: 8 822nd Pass: Normal Normal mode mode Shadow leak artifacts are completely removed WSCG‘03

Ming Li

Extended Opaque Projective Texture Mapping

Texture Texture-though Texture-though

WSCG‘03

Remove texture -through Remove texturetexture-through

INFORMATIK

Remove Remove Shadow Shadow leak leak

Ming Li

Rendering Results

INFORMATIK

Performance measurement •• •• •• ••

Triangle Triangle number number of of dynamic dynamic object object — — 400 400 ~~ 500 500 Texture -37% Texture packing packing improves improves rendering rendering speed speed by by 20 20-37% 15 15 fps fps for for visual visual hull hull reconstruction reconstruction 23 23 fps fps for for textured textured visual visual hull hull rendering rendering with with background background

WSCG‘03

Ming Li

Video

WSCG‘03

INFORMATIK

Ming Li

Conclusions & Future Work

INFORMATIK

Conclusions •• ••

Introduced Introduced aa novel novel blending blending algorithm algorithm to to hide hide boundary boundary seams seams for for background background rendering rendering Extended Extended opaque opaque projective projective texture texture mapping mapping to to handle handle shadow shadow leak leak artifacts artifacts

Future Work ••

WSCG‘03

Better Better and and faster faster rendering rendering by by exploiting exploiting flexible flexible programmability -generation graphics programmability of of next next-generation graphics hardware hardware Ming Li

Online Accelerated Rendering Visual Hulls in Real ...

Feb 6, 2003 - Online Accelerated Rendering of. Visual Hulls ... communication and sports analysis etc. Visual hull .... INFORMATIK. Ming Li. WSCG'03. Video ...

513KB Sizes 0 Downloads 110 Views

Recommend Documents

Online Accelerated Rendering of Visual Hulls in Real ...
Feb 7, 2003 - This paper presents an online system which is capable of reconstructing and rendering dynamic objects in real .... model are textured by different number of textures, we ..... Constructing virtual worlds using dense stereo.

Hardware-Accelerated Rendering of Photo Hulls -
Compared to a pure software implementation, the performance of our hardware-accelerated method .... OMR: Online Model Reconstruction for Interactive Virtual. Environments [Lok01]. ..... a P4 1.7GHz processor machine equipped with a GeForce. FX 5800 U

Hardware-Accelerated Rendering of Photo Hulls
Photo-consistency check without considering visibility ... Consider visibilities during photo-consistency check ... Constrain the area of each slicing plane. Reduce ...

Hardware-Accelerated Visual Hull Reconstruction and Rendering Talk ...
Jun 11, 2003 - Talk outline. Introduction. Previous ... Talk Outline. Introduction ... Key idea. • Perform 3D intersections on rasterized silhouette faces. Accelerate ...

Read eBook Real-Time Rendering, Third Edition Full ...
... to the book s contents new techniques worthwhile websites etc After coverage ... com News analysis and research for business technology professionals plus ...

siggraph-2016-limberger-poster--real-time-rendering-of-high-quality ...
D-14482 Potsdam, Germany www.hpi3d.de. Page 1. siggraph-2016-limberger-poster--real-time-rendering-of-high-quality-effects-using-multi-frame-sampling.pdf.

Real-time realistic rendering of clouds
These markers can be of many types for which the acquisition system records ... Figure 2.5: Tracking a human in motion using an infrared optical system.

Hardware-Accelerated Visual Hull Reconstruction and ...
Jun 11, 2003 - System Performance & Demo Video. Conclusions & Future Work. Introduction ... 3D interactive TV, computer games, tele-communication, etc.