c rowd p atches : p opulating l arge - s cale v irtual e nvironments for r eal -t ime a pplications...

CROWD PATCHES: POPULATING LARGE- SCALE VIRTUAL ENVIRONMENTS FOR REAL-TIMEAPPLICATIONSBarbara Yersin, Jonathan Maïm ,

Julien Pettré , Daniel Thalmann

(2009)

OUTLİNE

Introduction Related Work Crowd Patches Results Conclusions and Assesments

INTRODUCTİON

Crowd Patches approach Large-scale virtual environments Ability to generate in real-time Handle high density Low cost navigation simulation High motion quality

INTRODUCTİON

Create small areas – patches Motion trajectories Static and dynamic objects

Assemble patches to generate the environment

Templates to design the crowd patches

RELATED WORK

Global level Cell decompositions, voronoi diagrams Interactively handle large crowds

Local level Ruled based, social forces, potential fields

Lee et al. 2006 – Motion Patches

PATCHES

Endogenous objects (left) Periodic animation inside a patch

Exogenous objects (right) Periodic motions, from one patch to another

PATTERNS

Faces with input/output information

PATCH CREATİON

Assembly the patterns One for each face

Get the information on static and endogenous objects

Linearly initialize trajectories Make adaptations based on velocity Collision avoidance and smoothing

Particle based Attraction points and repulsions

PATCH CREATİON

ENVİRONMENT CREATİON

Bottom-Up Approach Iteratively create and add patches (top)

Top-Down Approach Decompose the geometric model of the

environment into cells (bottom)

PATCH TEMPLATES

To aid the designer Generate patches from given templates

Templates formed by: Static objects Endogenous objects Pattern types

PATTERN TYPES

Empty patterns No I/O – Represent borders with static obstacles

One-way patterns Inputs or outpus – One way flows

Specific I/O space distribution Limit the space of I/O – Narrow passages,lanes

Specific I/O time distribution Limit the time of I/O – Crossings sidewalks at

certain time

RESULTS

Bottom-up approach (without rendering)

Top-down approach 20 fps

Video!

CONCLUSİONS

Handle densely crowded large scale virtual environments

Realistic motion

Patches have identical time period Static patches Lacks variety Lacks interactivity

ASSESMENTS

Well constructed and written Clearly explains:

Objectives Motivation Contribution

Discusses: Shortcomings of the method Alternatives Future work

ASSESMENTS

Missing implementation details

Experiments? Only 2 applications are considered

Results Only result data is in terms of FPS

ASSESMENTS

Performance Reasonable top-down approach results Bottom-up would be slow

Real-time applications - limited Lack of dynamic updates Lack of goal queries

ASSESMENTS

Visually convincing (+) Motions and animations

Large scale virtual environments (+) On-line generation of the environment

Densely crowded (-) Many characters but not really dense Narrow passages, doorways?

ASSESMENTS

High quality/realistic motion (-) Avoidance is relatively poor Another fast approach might be integrated

Prediction based?

Collisions and pedestrian representations What if the trajectories in one patch are close

enough? Clearance, personal space Maybe simple disk representations

QUESTİONS & REMARKS

VR-Lab http://vrlab.epfl.ch/