3D Graphics 2 Simulation Engines 2008 Chalmers University of - - PowerPoint PPT Presentation

08-11-19 Simulation Engines 2008, Markus Larsson 1

3D Graphics 2

Simulation Engines 2008 Chalmers University of Technology

Markus Larsson markus.larsson@slxgames.com

08-11-19 Simulation Engines 2008, Markus Larsson 2

Camera management

 Three basic kinds of cameras in games

 First-person

 Camera attached to the player and inherits the exact

motion of the player

 Examples: Farcry, Doom, Quake, etc

 Scripted

 Camera moves along pre-defined paths  Examples: Alone in the dark, Resident Evil

 Third person

 The camera is located outside the body of the player and

shows both the avatar and the environment

 Examples: Super Mario 64, Gears of War

08-11-19 Simulation Engines 2008, Markus Larsson 3

Third-person camera: Constraints



The camera should never be closer to a wall than the near plane



It should never go outside a level



It should translate and rotate smoothly to always try to stay at a specific point in relation to the player character



It should smooth out discontinuities in the character's movement



It should be tested for collision detection



It should be able to enter the character when needed

08-11-19 Simulation Engines 2008, Markus Larsson 4

Third-person camera: Algorithm



Calculate the destination point for the camera from the character's position



The destination point is calculated by applying a displacement and rotation from the position of the player character. Different camera views can have different displacements, and it may be possible to switch between them.



Check the validity of the destination point (it could be on the wrong side of a wall)



Perform a ray intersect between character and destination point (there should be no intersection with the world geometry)



If the point is invalid, move the camera back towards the character so that it is positioned on the correct side of the wall



Calculate approximate translation and rotation motion and animate it over a series of frames (animation speed is a tweakable constant)



Check for collision using the bounding box of the camera during the motion

08-11-19 Simulation Engines 2008, Markus Larsson 5

Shaders and shader languages

 1990s

 Development of hardware for rendering of textured

3D primitives

 Hardware T&L introduced  Dynamic lighting done using Gouraud shading

 2000s

 Per-pixel shading  Real-time procedural textures  Advanced texture mapping techniques

08-11-19 Simulation Engines 2008, Markus Larsson 6

In pursuit of realism

 Traditionally, research on realistic computer

graphics has focused on global illumination methods such as ray tracing and radiosity

 Do not work in real time

 Pixar introduced the concept of shaders in

RenderMan and showed that GI is not strictly necessary for realistic images

 Instead, shader-based methods based on local

reflections can be used

08-11-19 Simulation Engines 2008, Markus Larsson 7

What is a shader?

 Three different interpretations from Watt & Policarpo, 2003  C-style module in the RenderMan API used for high-

level control of rendering components (surface, volume and light shaders)

 A combination of render states and texture maps for a

multi-pass or multi-texture render of an object on fixed- pipeline GPUs

 New hardware functionality for controlling the rendering

• f primitives on a per-pixel or per-vertex level on

programmable-pipeline GPUs; these are called pixel shaders and vertex shaders, respectively

 The last one is important one

08-11-19 Simulation Engines 2008, Markus Larsson 8

Shader types



Vertex shader



Called for every vertex in a 3D primitive



Allows for effects such as hardware skinning, perturbation of water surface, etc



Pixel (fragment) shader



Called once for every fragment in a 3D primitive (not pixel, because a fragment in a 3D primitive could correspond to one or several pixels depending on filtering settings, etc)



Can be used for procedural texture, normal maps, etc



Geometry shader



Can add to and remove vertices from a mesh and be used for adding geometry too costly to process on the CPU



Allows displacement mapping, etc



Unified Shader Model in DirectX 10 (Shader Model 4.0)

08-11-19 Simulation Engines 2008, Markus Larsson 9

Shader languages

 OpenGL Shading Language (GLSL)

 Part of the OpenGL specification since OpenGL 1.4  High-level language similar to C/C++

 Cg (C for Graphics)

 Nvidia's proprietary shader language  Extremely similar to HLSL, but works on both

OpenGL and DirectX

 Microsoft HLSL

 Works on DirectX 9 and 10  Nvidia and Microsoft collaborated on its

development

08-11-19 Simulation Engines 2008, Markus Larsson 10

Reflective surfaces: Environment maps

 Commonly used for reflections  Precomputed textures

 Standard environment maps

 Single texture representing the

scene in one texture as if reflected from a steel ball

 Cubic environment maps

 Cube map consisting of six textures unfolded on to a

cube

 The most used format on modern hardware

 Can also be used for advanced lighting

08-11-19 Simulation Engines 2008, Markus Larsson 11

Particle systems

 One of the most useful tools available

 Smoke, water, fire, etc

 Each individual particle has a small or zero

geometrical extent, but together form cloud-like

• bjects

 Each particle system can

contain thousands or even tens of thousands of particles

 Be extremely cautious of

multiple render states

08-11-19 Simulation Engines 2008, Markus Larsson 12

Particle systems



Consists of one or several emitters and a number of particles



Each emitter is responsible for spawning new particles every time update according to some distribution



Each particle contains information about its current position, size, velocity, shape and lifetime



Each update

 Emitters generate new particles  New particles are assigned initial attributes  All new particles are injected into the particle system  Any particles that have exceeded their lifetime are

extinguished (and usually recycled)

 The current particles are updated according to their scripts  The current particles are rendered

08-11-19 Simulation Engines 2008, Markus Larsson 13

Water



Ideally based on fluid simulations

 Way to costly for real-time use 

Mostly represented by a simple plane/quad

 In more advanced scenarios

represented by a displaced grid



Reflections are either entirely faked or by using a planar mirror

 The scene is inverted and the reflection is rendered to a

texture which is then rendered using projective texturing

• nto the water mesh

08-11-19 Simulation Engines 2008, Markus Larsson 14

Explosions



There is no universal method



Usually a combination of effects are used in combination until it “looks good”



Often animated billboards of prerecorded explosions are combined with debris that is either actual geometry or particles



Nowadays, often actual models are used which is moved by the physics engine

08-11-19 Simulation Engines 2008, Markus Larsson 15

Lightmaps

 Precomputed lighting and shadows on static geometry in

the scene

 Advantages  Allows for baking advanced lighting such as radiosity  Extremely fast  Disadvantages  Only works on static

geometry

 Requires a lot of pre-

computing for good quality

 May require a lot of

memory

08-11-19 Simulation Engines 2008, Markus Larsson 16

Stencil shadows



Used with dynamic shadows



Uses the stencil buffer



Advantages

 Creates crisp shadows without

aliasing artifacts

 Stable and potentially very fast algorithm 

Disadvantages

 Difficult to get soft shadows (but possible)  Extruding the shadow volumes on hardware is cumbersome

and requires modifications to the meshes

 Requires multiple render passes  Patent problems with Carmack's reverse

08-11-19 Simulation Engines 2008, Markus Larsson 17

Stencil shadows



Empty the stencil buffer



Draw the whole scene with ambient lighting

 The z-buffer is filled and the color buffer is filled with the

color of surfaces in shadow



Turn off updates to the z-buffer and color buffer and draw the front-facing polygons of the shadow volumes

 Increments the stencil buffer; all pixels in or behind shadow

volumes receive a positive value



Repeat for back-facing polygons

 Decrement the stencil buffer; the values in the stencil buffer

will decrease where we leave the shadow volumes



Draw the diffuse and specular materials in the scene where the value of the stencil buffer is zero

08-11-19 Simulation Engines 2008, Markus Larsson 18

Shadow maps



Texture based shadows



Renders the scene from the light's point of view



For each fragment, checks if it is the the closest

• ne to the light



Advantages



Can be done almost entirely on the GPU



Works very similar to drawing the scene regularly and allows reuse of frustum check code, etc



Good candidate for soft shadow algorithms



Disadvantages



Precision problems and aliasing artifacts



Good for directional and spot lights but omni lights may require up to six individual textures per light

08-11-19 Simulation Engines 2008, Markus Larsson 19

Soft shadows

 There are lots of techniques, but usually

shadow maps with multiple samples per texel are used

 Often extremely expensive to compute  “Free” in lightmaps

08-11-19 Simulation Engines 2008, Markus Larsson 20

Textures

 Texture coordinates (U,V)

 Unfolds the model

 Paint directly on the

model with software such as BodyPaint, Zbrush, etc

 Texturespace/

Tangentspace can be interesting for advanced effects such as Normal mapping

08-11-19 Simulation Engines 2008, Markus Larsson 21

Displacement mapping



Commonly used technique in raytracers



Not possible on DirectX 9 hardware

 Can not move vertexes that do

not exist

 Can be hacked for certain uses

by doing texture lookups in the vertex shader



Possible for real on DirectX 10 hardware

 Geometry shaders allow

tessellation of objects without CPU intervention

08-11-19 Simulation Engines 2008, Markus Larsson 22

Normal maps

 High-poly surfaces are transferred to

models by storing normals in textures

 Normals are normally stored in

tangent space

 Can produce extremely good results

with fairly low computational costs

 Works well on most types of objects lit

with dynamic lights

 Often, tools like Zbrush are used to

create high-poly objects, but normal maps can also be created from heightmaps

08-11-19 Simulation Engines 2008, Markus Larsson 23

Tangent space

08-11-19 Simulation Engines 2008, Markus Larsson 24

Parallax mapping

 Also know as offset mapping as it offsets the texture

coordinates

 Very cheap and fairly convincing on flat surfaces  High quality versions require an iterative process, which is

more expensive, but produce extremely good results (where applicable)

08-11-19 Simulation Engines 2008, Markus Larsson 25

Cel shading

 Non photo-realistic technique  Can be done either by drawing multiple passes

• r in a shader by examining the scalar product
• f the normal and view direction

 Check the sample in the Ogre SDK

08-11-19 Simulation Engines 2008, Markus Larsson 26

High Dynamic Range rendering

 Often done by cheating  Can be done for real on Shader Model 3.0 hardware  Render to a target with higher precision than can be

drawn

 Then scale the values in a second pass to simulate

different levels of exposure

08-11-19 Simulation Engines 2008, Markus Larsson 27

PRT



Precomputed radiance transfer can be used to simulate soft shadows, interreflections and subsurface scattering



Uses spherical harmonics to encode and approximate the light transfer function

 Decoded in a programmable shader  Only works on low frequency effects

with a reasonable number of parameters



Lights can move, but objects can not move relative to each other



Check the DirectX SDK

08-11-19 Simulation Engines 2008, Markus Larsson 28

Post processing

 Generally done by first rendering the scene to a

texture and then applying that texture to a full- screen quad which is drawn with a pixel shader applied to it

 Multiple post-processing effects can be

combined either in sequence or by blending between them

08-11-19 Simulation Engines 2008, Markus Larsson 29

Bloom

 Can be done best when combined with HDR rendering  In essence: apply gaussian blur on areas with intense light  A cheap substitute for actual blur is to downsample the

image a few times and then combining the original image with the downsampled versions

08-11-19 Simulation Engines 2008, Markus Larsson 30

Depth of Field



DOF is extremely important in real world photography and film



The modern approach (Thorsten Sheuermann, 2004)

 Downsample and pre-blur

the image

 Use variable size filter

kernel to approximate circle of confusion

 Blend between original

and pre-blurred image for better image quality

 Take measures to prevent

“leaking” sharp foreground into blurry background

08-11-19 Simulation Engines 2008, Markus Larsson 31

Motion blur



The lack of motion blur is mostly why 25-30fps in a game seems stuttering but works with film



The simplest form of motion blur is done by using the accumulation buffer and combining previous frames with the current

 Does not work well without very

high framerates



When used on meshes directly, a vertex shader stretches some vertices along the direction of motion and fades their transparency



Can be done in post processing by first drawing the scene to a velocity buffer which determines how much to blur the scene in the post processing pass

08-11-19 Simulation Engines 2008, Markus Larsson 32

Materials in Ogre

material Material/SOLID/TEX/sandstone.jpg { technique { pass { diffuse 0.8 0.8 0.8 specular 0.5 0.5 0.5 12.5 texture_unit { texture sandstone.jpg } } } }

08-11-19 Simulation Engines 2008, Markus Larsson 33

Post processing in Ogre

 There are plenty of pre-defined post processing

effects

 Compositor allows for easy access to the

effects

 Use compositor scripts to add effects or

combine multiple effects

 Check the Ogre SDK for demos

08-11-19 Simulation Engines 2008, Markus Larsson 34

Shadows in Ogre

 Extremely easy



mSceneMgr->setShadowTechnique(“techniquename”);

 Stencil shadows

 SHADOWTYPE_STENCIL_MODULATIVE  SHADOWTYPE_STENCIL_ADDITIVE

 Shadow maps

 SHADOWTYPE_TEXTURE_MODULATIVE  SHADOWTYPE_TEXTURE_ADDITIVE

 There are also more advanced methods

 Check the example in the SDK

08-11-19 Simulation Engines 2008, Markus Larsson 35

Next lecture

 Monday, 24 November  Artificial Intelligence