Advanced Game Technology CMPCD3026 CMPSEM044

1
Advanced Game TechnologyCMPCD3026-CMPSEM044
Abdennour El Rhalibi Room 723 a.elrhalibi_at_livjm.ac
.uk
2
Course Details (Attempt)

• 3D Game Engines Components
• DirectX D3D, 3D Modelling and Rendering
• Meshes, Level Loading and Editing
• Terrain Rendering and LOD
• Camera Setting and Animation
• Spatial Data structure BSP and PVS
• NPC Behaviour and 3D PathFinding A, Flocking,
  Scripting
• 3D Collision Detection and Response
• Shading languages
• Game networking Issues Architecture, Protocol,
  Event Synchronisation
• Introduction to Console Programming

3
Terrain Rendering and LOD Lecture 4

• Abdennour El Rhalibi

4
Terrain

• Terrain is important to many games
• As a model, it is very large
• Creating every point explicitly by hand is not
  feasible, so automated terrain generation methods
  are common
• When rendering, some of the terrain is close, and
  other parts are far away, leading to terrain LOD
  algorithms

5
Terrain

• A terrain mesh is nothing more than a triangle
  grid, with has heights of each vertex in the grid
  specified in such a way that the grid models a
  smooth transition from mountain to valley,
  simulating a natural terrain.
• To make it realistic we apply a nice texture
  showing sandy beaches, grassy hills, and snowy
  mountains

6
Terrain

• To generate terrain a simple solution consists in
  using a brute force approach.
• it simply stores the entire terrain vertex/index
  data and then renders it.
• For games requiring a small terrain, this
  approach is workable with modern graphics cards
  that support hardware vertex processing.
• However, for games requiring larger terrains, you
  have to do some kind of level of detail or
  culling because the enormous amount of geometry
  data needed to model such huge terrains can be
  overwhelming for a brute force approach.

7
What is the Problem?

• Terrains tend to be huge
• Visualizing a terrain of 16384 x 16384 samples
  (164 x 164 km, samples 10 m apart) requires
  drawing 536.805.378 triangles (268.435.456
  vertices)
• Adding a 32 bit RGBA texture and having 16 bit
  heights, the total memory consumption is above
  1.5 Gb!

8
Examples of Terrain with LOD
9
Examples of Terrain with LOD
10
Terrain Generation Methods

• Paint the height field (artist generated)
• Various pseudo-random processes
• Independent random height at each point
• Fault generation methods (based on random lines)
• Particle deposition
• Fractal terrain from meshes
• Triangulated point sample methods
• Plenty of research and commercial tools for
  terrain generation
• http//www.vterrain.org/LOD/Implementations/

11
Surface Attributes

• Rather than painting texture and color directly,
  paint attributes
• E.g. Grass, Trees, Water,
• Features can have game-play characteristics
  speed of motion, impassable,
• Then automatically generate colors/textures
• Care must be taken at the boundaries of the
  regions
• Can also work for the terrain itself
• E.g. Maps that have a finite number of possible
  heights, with ramps between them

12
Terrain Representation

• Terrains are often represented using elevation
  maps
• An elevation map is a 2D array of regularly
  spaced height samples

13
Painted Terrain Example
Color
Texture
Height
14
Representing Terrain

• The base representation for terrain is usually a
  height field
• zf(x,y) for (x,y) within the limits of the space
• There are two common ways to represent the
  function f(x,y)
• Explicitly store the value of f(x,y) for a
  discrete grid of (x,y) locations
• Generally interpolate or triangulate to get
  points not on the grid
• Easy to figure out what the height of the terrain
  is at any given (x,y)
• Expensive to store the entire terrain
• Store a polygonal mesh
• Cheaper if the terrain has large flat areas
• Harder to figure out what the height is under the
  player (have to know which triangle they are in)

15
Heightmaps

• We use a heightmap to describe the hills and
  valleys of the terrain.
• A heightmap is an array where each element
  specifies the height of a particular vertex in
  the terrain grid.
• One of the possible graphical representations of
  a heightmap is a grayscale map, where darker
  values reflect portions of the terrain with low
  altitudes and whiter values reflect portions of
  the terrain with higher altitudes.

16
Create Heightmap as Grayscale

• Heightmaps can be generated either
• procedurally
• or
• In an image editor such as Adobe Photoshop.
• Using an image editor is probably the easiest way
  to go, and it allows you to create the terrain
  interactively and visually as you want it.
• In addition, you can take advantage of your image
  editor features, such as filters, to create
  interesting heightmaps.

17
Random Processes Generation

• The claim is that real terrain looks random
  over many scales
• Hence, random processes should generate realistic
  terrain
• The catch is knowing which random processes
• Some advantages
• Lots of terrain with almost no effort
• If the random values are repeatable, the terrain
  can be generated on the fly, which saves on
  storage
• Some disadvantages
• Very poor control over the outcome
• Non-intuitive parameter settings

18
Random Points

• Randomly choose a value for each grid point
• Can make each point independent (dont consider
  neighbors)
• Can make points dependent on previous points - a
  spatial Markov process
• Generally must smooth the resulting terrain
• Various smoothing filters could be used
• Advantage Relatively easy to generate on the fly

19
Fault Formation

• Claimed to model real fault lines, but not at all
  like real faults
• Process 1
• Generate a random line and lift everything on one
  side of the line by d
• Scale d and repeat
• Process 2
• Generate a random line and lift the terrain
  adjacent to the line
• Repeat (maybe with a scale function)
• Smoothing is important

20
Fault Formation Example
Initial
Smoothed
21
Fault Formation Example
22
Particle Deposition

• Supposed to model lava flow (or glacial
  drumlins!)
• Process
• Drop a particle onto the terrain
• Jiggle it around until it is at most one unit
  higher than each of its neighbors
• Repeat
• Occasionally move the drop point around
• To do volcanoes, invert everything above a plane
• To do sinkholes, invert the hill
• To do rows of hills, move the drop point along a
  line

23
Particle Deposition Process
?
?
In 3D, more scope for random choices on jiggling
particles
24
Particle Deposition Image
25
Fractal Terrain

• Based on subdivision of a course polygon mesh
• Each subdivision adds detail to the mesh in a
  random way
• Algorithm (starting with a triangular mesh)
• Split each edge, and shift the new vertex up or
  down by a random amount
• Subdivide the triangles using the new vertices
• Repeat
• Also algorithms for quadrilateral meshes

26
Subdivision Method No 1
Note Works on any triangular mesh - does not
have to be regular or have equal sized triangles.
27
Subdivision Method No 2

• Generates a triangle bintree from the top down
• Useful for LOD,
• Ideally, works for right-angled isosceles
  triangles

28
Subdivision Method No 3

• Assume quadrilateral meshes

29
Fractal Terrain Details

• The original mesh vertices dont move, so it
  defines the overall shape of the terrain
  (mountains, valleys, etc)
• There are options for choosing where to move the
  new vertices
• Uniform random offset
• Normally distributed offset small motions more
  likely
• Procedural rule eg Perlin noise
• making patterns from pseudo-random numbers
• If desired, boundary vertices can be left
  unmoved, to maintain the boundary edge

30
Fractal Terrains
Very jagged terrain
http//members.aol.com/maksoy/vistfrac/sunset.htm
31
Terrain, clouds generated using procedural
textures and Perlin noise http//www.planetside.co
.uk/ -- tool is called Terragen
32
Populating Terrain

• Coloring terrain
• Paint texture maps
• Base color on height (with some randomness)
• Trees
• Paint densities, or randomly set density
• Then place trees randomly within regions
  according to density
• Rivers (and lakes)
• Trace local minima, and estimate catchment areas

33
Terrain Generation Trade-Offs

• Control vs Automation
• Painting gives most control
• Fractal terrain next best control because you can
  always specify more points
• Random methods give little control - generate
  lots and choose the one you like
• Generate on-the-fly
• Random points and fractal terrain could be
  generated on the fly, but fractal terrain
  generation is quite slow
• Tilings can also be generated on the fly

34
Static LOD

• Depending on the roughness of the terrain and the
  application, 5-50 of the vertices and triangles
  can be removed
• With 536.805.378 triangles still more than
  200.000.000 triangles to draw in best case.
• Frustum culling further reduces number of
  triangles to draw
• In most cases we still draw the terrain at full
  resolution near the far plane

35
View-Dependent Dynamic LOD

• Dynamic simplification of visible part of the
  terrain
• A mountain observed from a distance of 10 km
  requires a higher tessellation than when observed
  from a distance of 100 km
• The quality of the tessellation can be changed at
  run time to achieve constant frame rates
• Terrains can be altered at run time

36
Terrain LOD Algorithms

• Triangle bintree based
• ROAMing Terrain Real-time Optimally Adapting
  Meshes Duchaineau et al.
• Quad tree based
• E.g. Real-Time, Continuous Level of Detail
  Rendering of Height Fields Lindstrom et al.
• Progressive mesh based
• E.g. Smooth View-Dependent Level-of-Detail
  Control and its Application to Terrain Rendering
  Hoppe
• Geo Mipmapping
• Fast Terrain Rendering Using Geometrical
  MipMapping de Boer

37
Terrain LOD

• Terrain poses problems for static LOD methods
• Must have high resolution in the near field, and
  low resolution in the distance, all in one model
• Dynamic LOD methods are the answer
• All based on the idea of cuts through a tree of
  potential simplifications
• We will discuss the ROAM algorithm in detail
• Other continuous LOD algorithms are similar in
  style
• An alternative is to create fixed LODs for
  sub-regions and figure out how to join them at
  the seams

38
Making Terrain LOD Easier!

• Assumption We are starting with a height field
  defined on a regular grid
• Assume its a square to make it easier
• We can mesh it by forming triangles with the data
  points
• The data is highly structured
• Every data point has the same number of neighbors
• Every triangle can be the same size
• Hence, the tree of possible simplifications is
  very regular
• Still, multiple possibilities exist for the
  triangulation and the simplification operations

39
Triangle Bintrees

• Binary trees in which
• Each node represents a right-angled isosceles
  triangle
• Each node has two children formed by splitting
  from the right angle vertex to the midpoint of
  the baseline
• The leaf nodes use vertices from the original
  height field
• Another way to build a spatial partitioning tree,
  but particularly well suited to simplification
  algorithms
• Easy to maintain neighbor information

40
Triangle BinTrees

• A triangle bintree is a binary tree, whose nodes
  consist of right-isosceles triangles
• the children of a node are defined by splitting
  the triangle along an edge from the apex vertex
  to the midpoint of the opposite edge

41
Triangle Bintree Example
3
1
4
6
2
1
2
5
3
4
5
6
8
7
7
8
9
10
11
12
13
14
10
13
14
9
11
12
42
Bintree Data Structure

• Parent and child pointers
• Neighbors
• A left neighbor, a right neighbor, and a base
  neighbor
• Note that the base and right angle give us a way
  to orient the triangle
• Neighbors are not necessarily at your own level
• Later, error bounds that say how much variation
  in height there is in your children

43
Triangulation

• Triangle bintrees form a mesh by assigning a
  world space position to each vertex in the tree
• Higher (lower) levels of detail are obtained by
  splitting (merging) triangles
• Adds up to traversing the tree one level further
  down
• Care must be taken to avoid crack in the
  resulting mesh

44
Avoiding Cracks

• Neighbor triangles are always from same level,
  one finer (left right) or one coarser (base)
• When splitting a triangle, its base neighbor is
  split at the same time
• If the base neighbor is at a coarser level, it
  must be split first, which may require further
  recursive splitting

45
Generating Cuts

• Cuts are generated by a sequence of split or
  merge steps
• Split Drop the cut below to include your
  children
• Merge Lift the cut up above two children
• To avoid cracks, some splits lead to other,
  forced, splits
• An LOD algorithm chooses which steps to apply to
  generate a particular triangle count or error rate

46
Cuts
3
1
4
6
2
1
2
5
3
4
5
6
8
7
7
8
9
10
11
12
13
14
10
9
47
Neighbors

• 5 left neighbor 6, right neighbor 9
• 6 left neighbor 8, right neighbor 5
• 7 left neighbor 8, base neighbor 10
• 8 base neighbor 6, right neighbor 7
• 9 base neighbor 5, left neighbor 10
• 10 base neighbor 7, right neighbor 9

8
7
10
6
9
5

• Note that 8 is 6s left neighbor but 6 is 8s
  base neighbor
• If you are someones left/right/base neighbor
  they are not always your right/left/base neighbor
• In other words, neighbors need not come from the
  same level in the tree

48
Not All Cuts Are Created Equal
3
1
2
1
4
6
2
3
4
5
6
5
7
8
9
10
11
12
13
14
8
7
10
Note the T-vertex - causes cracks in rendering
9
49
A Split

• A split cuts a triangle in two by splitting its
  base edge
• If the base edge is on a boundary, just split, as
  shown
• If the base edge is shared, additional splits are
  forced
• Add a new triangle to the mesh

2
1
3
4
5
6
6
7
8
9
10
11
12
13
14
50
Forced Splits

• Triangles are always split along their base
• Hence, must also be able to split the base
  neighbor
• Requires neighbors to be mutual base neighbors
• If they are not base neighbors, even more splits
  are needed
• Simple recursive formulation

51
Merges

• A diamond is a merge candidate if the children of
  its members are in the triangulation
• The children of the 7-10 diamond below are
  candidates
• Look for parents of sibling leaf nodes that are
  base neighbors or have no base neighbors
• Reduces the triangle count

8
7
10
9
52
Dual Queue Split-Merge

• Dynamic triangulation is driven by one or two
  priority queues
• Greedy split only algorithm uses a split queue to
  determine what triangles to split
• Incremental algorithm uses both a split queue and
  a merge queue to take advantage of frame-to-frame
  coherence
• Each triangle is given a priority
• Triangles with high priorities are split before
  triangles with low priority

53
Split Queue

• Insert all triangles in base triangulation into a
  queue
• While stopping criteria is not met
• Find and split triangle with highest priority
• Remove that triangle and other split triangles
  from the queue
• Add any new triangles to the queue

54
Split-Merge

• If this is the first frame Insert all triangles
  in base triangulation into the split queue
• Else Update priorities for all elements in the
  two queues
• While quality of triangulation not satisfactory
  or maximum split priority gt minimum merge
  priority
• If triangulation is to fine
• Find and merge lowest priority mergeable diamond
• Remove merged triangles from split queue and add
  the merge parents
• Remove diamond from merge queue and add new
  mergeable diamonds
• Else
• Find and split highest priority triangle from
  split queue
• Remove spilt triangles from split queue and add
  any new triangles
• Remove diamonds from merge queue, whose children
  were split
• Add any new mergeable diamonds to merge queue

55
Refinement LOD Algorithm

• Start with the base mesh
• Repeatedly split triangles until done
• Stop when a specific triangle count is reached,
  or
• Stop when error is below some amount
• To guide the split order, assign priorities to
  each split and always do the one with the highest
  priority
• After each split, update priorities of affected
  triangles
• Sample priority High priority to splits that
  will reduce big errors
• What is the complexity of this? (Roughly)
• A similar algorithm works by simplifying the mesh
  through merge operations. Why choose one over the
  other?

56
Performance Bottlenecks

• Storing and managing priorities for out-of-view
  triangles is a waste of time
• Do standard frustum culling to identify them
• Sending individual triangles is wasteful
• Build strips as triangles are split and merged
• Naively, at every frame, wedgies must be
  projected, new priorities computed and the queues
  re-sorted
• Use the viewers velocity to bound the number of
  frames before a priority could possibly make it
  to the top of a heap
• Delay recomputation until then
• Priority queue Bin priorities to reduce sorting
  cost
• At low priorities, order within bins doesnt
  matter

57
Additional Enhancements

• Stop processing after a certain amount of time
• Easily done just stop processing the next split
  or merge
• Result no longer optimal, but probably not bad
• Cost of dual queue algorithm depends on the
  number of steps required to change one mesh into
  another
• Check ahead of time how many steps might be
  required
• If too may, just rebuild mesh from scratch using
  refinement algorithm
• Can get accurate line-of-sight or under-vehicle
  height by manipulating priorities to force
  certain splits

58
Other Algorithms

• Algorithm from Lindstrom et al is essentially the
  same idea as the ROAM refinement algorithm
• Based on square blocks of (triangulated) terrain
• Makes determining forced splits harder
• Slightly different error bounds
• An algorithm from the game community
• For a particular split operation, the viewer
  distance is the primary error factor
• For each split, store distance at which it should
  occur
• Check current mesh on each frame for
  splits/merges according to viewer distance
• Non-optimal, but gets rid of priority queues (the
  big cost)

59
Other Issues

• Terrain Texturing
• Terrain Lighting
• Camera Animation and Fly-through
• SkyBox
• Collision
• Maintaining characters and objects on top of
  Terrain