1. BATTLECAM™: A Dynamic Camera System for Real-Time Strategy Games Yangli Hector Yee Graphics Programmer, Petroglyph hector@petroglyphgames.com Elie Arabian Lead Artist, Petroglyph elie@petroglyphgames.com

2. Overview  Background  Theory  Implementation  Hacks  Cinematic Shots  Question & Answers

3. Background – History  RTS Bird’s Eye (Dune 2)  First Person (Dungeon Keeper)  Scripted Actor & Camera (Warcraft 3)  Unscripted Actor, Unscripted Camera (Star Wars – Empire at War)

4. Background - Problem  Make a ‘movie’ from an RTS battle  Actors can move during shot  Actors can die during shot  Players can move actors  Objects can move into camera

5. Background - Solution  Pick most interesting object  Construct shot  Play shot  Fallback on death  Pick next object  Hijack existing camera scripting

6. Theory – Visual Attention  How to pick ‘interesting object’  Bottom Up: Stimulus  Intensity (black on white)  Motion (moving stuff)  Color (red on green)  Orientation (circle in stripes)  Top Down: Goals  Game Objectives  Current User Selection

7. Theory – Bottom Up Attn. Image Orientation Color Spatial Frequency Spatial Frequency Intensity Motion Feature Maps OrientationColor Spatial Frequency IntensityMotion Conspicuity Maps Saliency Map Reference : Itti L, Koch C. “A Saliency-Based Search Mechanism for Overt and Covert Shifts of Visual Attention.” Vision Research, pp. 263, Vol 40(10 - 12), 2000

8. Center Surround Mechanism Lateral Inhibition Intensity Feature Maps Intensity Conspicuity Maps

9. Lateral Inhibition One signal vs similar signals

10. Lateral Inhibition Purpose : Promote areas with significantly conspicuous features while suppressing those that are non-conspicuous. (Before Inhibition) (After Inhibition)

11. Implementation  Game Logic Driven  Images too expensive  No screen space stuff  Access to game logic info

12. Implementation – Data  Game logic data (stimulus)  Size  Attack power  Location  Health  Game logic data (goal driven)  Current selection  Visibility

13. Computing Saliency  E.g. Saliency_Speed for object(i)  Saliency_speed(i) = (speed(i) – min_speed)/ (max_speed – min_speed)  Normalized 0 to 1 Three Normalization modes Large is important Small is imporant Closeness to mean is important

14. Normalization Modes  Large is important  Saliency_val(i) = (val(i) – min_val) / (max_val – min_val)  Small is important  Saliency_val(i) = 1 – (val(i) – min_val) / (max_val – min_val)  Close to mean is important  Saliency_val(i) = 1 – (val(i) – avg_val) / (max_val – min_val)

15. Normalization settings  Large is important  Size  Attack power  Targets  Speed  Small is important  Health  Close to mean is important  X, Y coordinate

16. Lateral Inhibition  Conspicuity value = saliency_val * (max_saliency_val – min_saliency_val)  Signals with great difference between max and min get boosted

17. Importance  Importance (i) = Sum of conspicuity_vals * weights  Weight values  Size 1.0  Power 1.0  X 0.5  Y 0.5  Health 1.0  Targets 1.5  Speed 1.0  Sort list by importance

18. Summary  Compute normalized saliency  Perform lateral inhibition  Weighted sum  Sort by importance

19. Picking interesting object  Pick current selected  Pick current object’s target 50% of the time  Make interesting object list  From list pick top 5 randomly. Reject if it was same type as the previous object looked at.

20. Constructing Cinematics  Local Space  Transform object space cinematic into world  Local Space without rotation frame  Use translation only  World space using reference objects  For artist driven cinematic constructed in world space  Transform to local space

21. Local Space Cameras

22. Flyby camera shot

23. Circle camera shot

24. Chase camera shot

25. Hardpoint camera shot

26. Frigate/Target camera shot

27. Demo & Q&A  Thanks to  Jim Richmond for camera system  Kevin Prangley for illustrations  Petroglyph staff for support  Contact Info  Hector at petroglyphgames dot com