Nicolas Halper

1
ACameraEngine For Computer Games

• Nicolas Halper
• Ralf Helbing
• Thomas Strothotte

Otto-von-Guericke University of Magdeburg
2

• Introduction
• Camera in Game Pipeline
• The Predictive Camera Planner
• Results
• Conclusion

3
Introduction

• Every game is now 3D
• Presentation of 3D world influences playing
  experience
• Content is becoming more involved
• Story-telling capabilities
• Tournament Spectator modes

4
Previous Work

• Computer Games
• Pre-computed cut-scene
• Specialised camera routines
• Constraint Satisfaction
• Cost optimisations (Drucker 94, Bares 99)
• Robotics visual servoying (Marchand 00)
• Cinematography
• Virtual Cinematographer (He 96, Christianson
  96)
• Behavior-based automonous agents (Tomlinson 00)

5
Current Problems

• Computer Games
• Tailored camera agents for each game
• Often confuse the player
• Constraint-Solving approaches
• Problems with visibility constraints
• Frame-Coherency in camera motion not part of
  equation
• Cinematography
• Computer games are as different from Movies as
  Movies are from Theatre - Barwood
• Limited to specific scenarios (e.g. cut-scenes)

6
Motivation

• Long-term Goal
• Director chooses presentation for player
• High-level idioms built on low-level camera
  constraints
• Current Goal
• Fitting a Camera Engine into a game framework
• Heterogenous constraints to work together
• Achieve frame-coherent results in real-time

7
Game Pipeline
Flexible constraints
Player
8
Camera Module
Director
Camera Planner
Shot Template
Visibility Solver
parameterised constraints
unoccluded position
position
Emotion Template
Constraint Solver
Shot Library
solved camera state
camera state
request settings for time t
Camera Expert
Template Selector
events
next-frame camera state
Action
Lighting
9
Camera Module
Director
Camera Planner
Shot Template
Visibility Solver
parameterised constraints
unoccluded position
position
Emotion Template
Constraint Solver
Shot Library
solved camera state
camera state
request settings for time t
Camera Expert
Template Selector
events
next-frame camera state
Action
Lighting
10
Constraint Solver
11
Constraint Solver

• Constraint Set taken suitable for computer games
• Achieve certain desired shots
• Constraint-Solving pipeline
• Allows heterogenous constraints to be solved
  together

12
Constraint Solver

• For each constraint
• For each frame, compute goal camera state
• Alter the current camera state to converge
  towards the goal camera state
• Satisfy constraints within tolerance regions

13
(No Transcript)
14
Inside Tolerance region pos ratio (optimal
pos)
Outside Tolerance Region clamp to min/max
15
Inside Tolerance region pos ratio (optimal
pos)
Outside Tolerance Region clamp to min/max
16
Inside Tolerance region pos ratio (optimal
pos)
Outside Tolerance Region clamp to min/max
17
(involved process more on this later)
18
(No Transcript)
19
Visibility Solver
Camera Planner
Visibility Solver
unoccluded position
position
Constraint Solver
solved camera state
camera state
Camera Expert
20
Potential Visibility Regions(PVR)

• Method introduced to solve visibility constraints
• Flexible
• Robust
• Multiple points can be specified
• PVR defined by geometry
• Input at run-time
• Adapt to constraints

21
PVR
If camera cant see the target, target cant see
the camera
Where can we move the camera, so that it can
see the target?
target
Select the best colour-coded region
22
PVR Multiple Targets
Target A
Target B
23
PVR Multiple Targets
Target A
Target B
24
PVR Flexibility
PVR constrained to .
25
Camera Expert
Camera Planner
Visibility Solver
unoccluded position
position
Constraint Solver
solved camera state
camera state
request settings for time t
Camera Expert
26
Camera Expert

• Satisfy visual goals and maintain frame-coherence
• Estimate future conditions
• Adapt camera trajectory appropriately

27
Camera Expert
Goal is to find next frame state for the
camera
past cameras
past targets
28
Camera Expert
Goal is to find next frame state for the
camera
First predict according to past camera motion
what state camera will be in at time t in future
29
Camera Expert
First predict according to past camera motion
what state camera will be in at time t in future
Also predict target state at time t in future
past cameras
past targets
30
Camera Expert
First predict according to past camera motion
what state camera will be in at time t in future
Also predict target state at time t in future
past cameras
past targets
31
Camera Expert
past cameras
past targets
32
Camera Expert
Now compute expected camera position for
next frame such that at time t we will arrive at
FPSC
past cameras
past targets
33
Camera Expert
Now compute expected camera position for
next frame such that at time t we will arrive at
FPSC
past cameras
past targets
34
Camera Expert
Camera Expert works each frame to give smooth
motion over time
35
Results

• Subtle
• Attic Scene
• Highly complex spatial properties
• 50k polygons, each tested for occlusion
• City Scene
• 200k polygons
• Reduced 1.1k polygons for occlusion tests
• Medieval Scene
• Tightly enclosed environment
• Athletic character!

36
Conclusion

• Arbitrary dynamic scenes and spatial complexities
  of environments
• Efficient and highly flexible visibility
  constraints using PVR
• Compute from existing state and motion
  characteristics to adapt to future conditions
• Combine heterogenous constraints in
  frame-coherent manner
• Results are fast, consistent, robust

37
Current and Future Work

• Spectator Mode
• Networked Tournament game
• Spectators follow the action
• Layed the ground work for Director
• Encoding some cinematic idioms
• Develop camera language suitable for interactive
  techniques