Transcript PAPER WRITING CLINIC

Perceiving Motion Transitions in
Pedestrian Crowds
Qin Gu, University of Houston
Chang Yun, University of Houston
Zhigang Deng, University of Houston
Virtual Reality Software and Technology (VRST) 2010
Introduction
Walking motions of real pedestrians vary in both spatial and temporal
domains. However, computer-generated pedestrians typically
repeat the same walking pattern all the time.
“Robotic” crowd
Real crowd
UH CGIM Lab
Related Work
Improving crowd motion variety
given a set of walking motion patterns:
1.
Randomly select motions
2.
Select motions based on examples
[LCHL07], [LCL07], [LFCC09]
3.
Select motions via heuristic rules
[PAB07], [YT07], [GD10],
[LFCC09] Fitting Behaviors to Pedestrian Simulations,
SCA 09
UH CGIM Lab
Motivation
1.
Interpolating motion patterns introduce unrealistic motion
transitions.
2.
Most transition optimizations for single character are
computation consuming. [RGBC96] [KGP02]
Our objective

how “macro” crowd features make an illusion that the
animation quality of each character in the crowd is visually
improved without utilizing sophisticated optimization
techniques.
UH CGIM Lab
Experiment Specifications






HiDAC model [PAB 07].
Strategy view & FPS view
36 student participants
38 trials with 20 seconds of each
Simple interpolation
Uniform motion transition rate
Crowd Density Effect
Strategy view
Density: 8
Density: 64
FPS view
Crowd Density Effect (2)

Two-way analysis of variance was used to evaluate the average
transition frequencies rated by the participants. (4 – 64 average density)

Main effects:
- Density of the crowd
(F = 12.89, p < 0.017)
- Viewpoint
(F = 32.91, p < 0.001)

Interaction:
(F = 15.76, p < 0.018)
Appearance Variety Effect
Strategy view
FPS view
1 texture
16 textures
UH CGIM Lab
Appearance Variety Effect (2)

Two-way analysis of variance was used to evaluate the average
transition frequencies rated by the participants. (1 – 16 textures)

Main effects:
- Appearance number
(F = 17.72, p < 0.014)
- Viewpoint
(F = 23.13, p < 0.008)

Interaction:
no evident interaction
UH CGIM Lab
Motion Variety Effect
Strategy view
FPS view
2 Motions
10 Motions
UH CGIM Lab
Motion Variety Effect (2)

Two-way analysis of variance was used to evaluate the average
transition frequencies rated by the participants. (2 – 10 motions)

Main effects:
- Motion number
(F = 17.72, p < 0.014)
- Viewpoint
(F = 37.76, p < 0.006)

Interaction:
no evident interaction
UH CGIM Lab
Group Interaction Effect
random
chase
flocking
advection
UH CGIM Lab
Group Interaction Effect (2)

Two-way analysis of variance was used to evaluate the average
transition frequencies rated by the participants. (4 interactions)

Main effects:
- Motion number
(F = 44.56, p < 0.004)
- Viewpoint
(F = 14.97, p < 0.012)
Interaction:
not available
UH CGIM Lab
Summary

A series of psychophysical experiments to investigate the influences
of different viewpoints, crowd densities, appearance variations,
motion variations, and collective group interactions.
- Strategy viewpoint is more effective to hide motion transitions
- Increasing the density of agent numbers helps to hide motion transitions.
- Adding agent appearances does not lead to better perception of motion
transitions in a crowd.
- Increasing the number of motion candidates makes motion transitions
easier to be detected
- Collective behaviors or sub-group interactions can effectively decrease the
negative impact of motion transitions.
UH CGIM Lab
Future work

Investigate the interactions among density, appearance variety and
motion variety.

Perform experiments on off-line crowds.

Probe the transition perceptions on other types of crowd motions,
such as running, talking, and fighting.
UH CGIM Lab
Thank you!
Project page:
http://graphics.cs.uh.edu/projects/CrowdTransitionPerception/

Presenter: Qin Gu
UH CGIM Lab
NSF IIS-0914965 & Texas NHARP 003652-0058-2007