Crowd Self-Organization, Streaming and Short Path Smoothing

學號： 9555535姓名：邱欣怡日期： 2007/1/2

Stylianou Soteris & Chrysanthou Yiorgos

Outline

Introduction Related work Method & Algorithm Result Performance Conclusion

Introduction

Collective behavior of pedestrians, known as self-organization

Flow grid mechanism in introduced to simplify navigation and enable crowd self organization

Related Work

Crowd navigation Deal with the problem of steering an

avatar inside a large amount of static and moving obstacles

Related Work (conti.)

Approaches to solving the navigation problem

1. Path planning method Not good for dense crowd navigation

2. Reactive navigation method Significant work

3. Behavioral navigation method Simulate aspects of pedestrian behavior

Related Work (conti.)

Reactive navigation method1. Force field method

Collision avoidance and smooth steering2. Rule based method

Model complex behaviors through a combination of attributes and rule, and through FSM

3. XZT space method Using space-time represent movements of d

ynamics objects

Methodology

First ,flow grid is constructed over the walk area Flow grid->measure densities and

velocities at various directions Using flow grid to navigate Using Steering algorithm to local

steering and smoothing

Measuring the Flows

Each avatar is registered on the grid by distributing his density and velocity to the 4 neighboring points

Velocities are separated into X and Z axis components and stored at each point, (+vx,-vx,+vz,-vz)

Complete Picture

Using the Flow Grid to Navigate

Weight = (1+D) * (1+AngleDiff(T,F)) D = density at spot T = vector showing direction towards target pos F = vector showing direction of flow at spot AngleDiff =Angle difference between 2 vectors in radians

Steering Algorithm

Using discrete occupancy map

Actual avatar First 4 position are used for curve interpolation (Catmul-Rom curve)

Last 2 position are used for path smoothing (in order to minimize the turn angle)

Next position search

Search starts from the top and checks left and right at an increasing angle and reducing distance until an empty cell found

Result

Two Parallel But Opposing Streams

Result

Two Crossing Crowd Streams

Result

Resulting paths of parallel but opposing streams

Resulting paths of perpendicularly crossing streams

Performance

Improved Density of avatars. Approximate crowd jam limits for different area sizes

Performance

Local Navigation performance cost

Conclusion

Flow grid gives an overall impression of what kind of pedestrian traffic exists in that area, and it can detect Congested araes.

Steering algorithm can reduce the collisions avoidance cost.