Rendering of Realistic Landscapes

1
Rendering of Realistic Landscapes

• K. H. Ko
• Department of Mechatronics
• Gwangju Institute of Science and Technology

October 29, 2008
2
Introduction

• Creating detailed three dimensional landscapes
  manually is a costly and slow process.
• Automatic generation of landscapes is needed.
• Features in a landscape
• Terrain with a varying topology and surface
  structure the basis
• Plants forests, fields
• Individual buildings, cities, roads,
  infrastructure
• Rivers, lakes, seas, sky, clouds, the sun, stars
• Animals and people, etc.

3
Introduction

• To create a realistic landscape
• It is not enough that the elements making up the
  landscape look realistic.
• They should also be placed naturally in relation
  to each other.
• Mimicking the way real landscapes are structured.
• Ecotopes provide a way to achieve this.
• A consistent landscape with variation both at
  local and global scales is more interesting than
  a completely random or homogeneous landscape.

4
Introduction

• Applications
• Flight simulators, computer games, visualization
  in architecture, land use planning tools,
  geographical visualization, landscape design,
  background generation for movies, etc.
• Dynamic and interactive landscapes but could be
  artificial
• Computer games
• Static but accurate landscape
• Geographical information system

5
Procedural vs. Declarative

• Declarative Approach
• Define every details and properties of every
  object in the landscape.
• The designer has absolute control over the
  landscape.
• It requires a huge amount of storage space.
• Creating a landscape takes time.
• It is used for real world geographical system.
• Conformance to the real world geography is
  important.
• The terrain height and object placement
  information can be obtained from measurement data.

6
Procedural vs. Declarative

• Procedural Approach
• It generates the landscape using algorithms that
  produce varying, natural looking data.
• This approach can be used if the landscape does
  not have to match any real life location.
• Almost no input data is required.
• Specialized algorithms are used.
• Small disk storage and minimum work

7
Hybrid

• Both the procedural and declarative methods are
  combined.
• Use the procedural approach by default
• Allow exact definitions in places
• The landscape designer can accurately specify
  details where they are needed.

8
Ecotopes

• Different areas in an extensive natural landscape
  may have very different appearances.
• Rocky ground, forests, lakes, etc.
• Ecotopes are a way to implement this kind of
  variation.
• Ecotopes provide a flexible framework that can be
  used to implement both macro and micro scale
  features
• Macro scale features forests
• Micro scale features tall grass
• They give landscape designers both controls over
  the characteristics of different types of
  landscape, as well as control over where to apply
  what type of landscape.

9
Ecotopes

• Landscape parameters of an ecotopes
• Height functions
• A number of different plant species and their
  densities
• Rain amounts which affect the number of rivers
  and lakes generated.
• Population density
• Number of buildings
• Etc.

10
Ecotopes

• Distribution Properties of Ecotopes
• Terrain elevation, relative elevation, slope
  angle, proximity to sea, proximity to a river or
  lake, etc.
• Randomly generated noise function

11
To Create a Landscape

• Terrain
• Plants
• Trees, grass, etc.
• Buildings
• Cities
• The Sky
• Clouds, weather, climate, atmosphere, celestial
  bodies, etc.

12
Terrain

• It is the basis for a landscape, i.e. the shape
  of the ground.
• For efficiency in modeling and rendering the
  ground, we may assume that
• The ground surface has no overhangs.
• Any ray from the center of the planet intersects
  the planet surface exactly once.
• Under this assumption, the ground shape can be
  defined by a ground height function.
• When modeling and rendering only a small part of
  the planet surface, we treat the surface as
  planar.

13
Terrain

• For rendering a ground surface, it is often
  practical to only store samples of the ground
  height function at some intervals.
• Called an elevation map or height map.
• If the height is encoded as colors, it can be
  stored as an image.

14
Random Terrain Generators

• The simplest way is to assign each position on
  the ground a random height.
• The result of that bears little resemblance to
  natural terrain.
• In fact the natural ground is more or less
  continuous, while still varying in height in
  complex ways depending on the position.

15
Random Terrain Generators

• Stochastic Subdivision Algorithms
• Iterative subdivision with pseudo-random midpoint
  displacement
• Algorithm
• The terrain starts with a single large square,
  with a height value of zero at each corner.
• A pseudo-random height offset that is
  proportional to the size of the square, is added
  to each corner of the square.
• The square is divided into four smaller ones,
  with the height of each new corner interpolated
  between the heights of neighboring corners of the
  original square.
• The algorithm is repeated from step 2 for each
  square, until the squares are at the desired LOD.

16
Random Terrain Generators
Height Map
3D Rendering
17
Random Terrain Generators

• Stochastic Subdivision Algorithms
• Drawbacks
• It often produces unnatural looking regularities
  (sharp ridges or peaks)
• The random variation varies linearly with the
  scale of the features.
• In nature terrain the amplitude of height
  variation does not depend linearly on the scale
  of the features.

18
Random Terrain Generators

• Stochastic Subdivision Algorithms
• Diamond Square Subdivision
• The algorithm divides a square into four smaller
  squares rotated 45 degrees in relation to the
  original square.
• It eliminates some of the more visible artifacts,
  but has some quite noticeable point-like
  artifacts of its own.

19
Random Terrain Generators

• Stochastic Subdivision Algorithms
• Offset square subdivision
• It can avoid most of the artifacts, but with
  somewhat increased performance cost.
• The smaller squares are offset from the larger
  square corners, and the initial values for the
  smaller square corners are calculated with a
  weighted average.
• More smooth terrain

20
Random Terrain Generators

• Faulting Algorithms
• They generate fractal data by repeatedly dividing
  the terrain with a faulting edge.
• Raising the terrain on one side of the edge and
  lowering it at the other to achieve a height
  difference along the faulting edge.
• Over time the height difference is reduced and
  when it arrives at zero the terrain is ready.

21
Random Terrain Generators

• Faulting Algorithms
• Very slow
• Not suitable for applications where a small
  visible area of a larger terrain needs to be
  generated.

22
Random Terrain Generators

• Perlin Noise
• It approximates smooth white noise of a given
  frequency in one or more dimensions.
• Combining several layers (called octaves) of
  noise at different frequencies and amplitudes, a
  natural looking fractal noise can be obtained.
• This combined noise is called Perlin turbulence,
  or just Perlin noise.

One layer
23
Random Terrain Generators

• Perlin Noise
• The characteristics of a terrain height field can
  be adjusted by changing the number of octaves,
  the amplitude and frequency of each octave.

24
Random Terrain Generators

• Perlin Noise
• It does not have any regular visible artifacts
  and is fast.
• It is a popular choice for random terrain
  generation.
• One variation of Perlin turbulence is the ridged
  multi-fractal noise.
• It uses absolute value functions to produce
  features with a ridged appearance.
• Approximate eroded mountain ranges.

25
Random Terrain Generators

• Successive Mass Deposit
• It is based on the idea of repeatedly adding some
  mass at a random location of the terrain.
• The mass has a Gaussian distribution profile
  around the addition point.
• The addition is repeated with successively
  smaller masses.

26
Geological Effects on the Terrain

• A height field that statistically resemble the
  real landscape topographies visually lacks many
  of the distinct geological features found in real
  world landscapes.
• The result of various geological processes
• Topology building processes
• Move the planet crust or allow magma to rise to
  the surface.
• Erosive processes
• Wear down the planet crust to progressively more
  fine grained particles.
• Transport these particles and deposit them in new
  places.

27
Terrain Level of Detail

• Level of Detail For Terrain Visualization
• Easier than arbitrary 3D models
• The geometry is more constrained, normally
  consisting of uniform grids of height values.
• More specialized and potentially simpler
  algorithms could be possible.
• More Difficult
• It is possible to have a large amount of terrain
  visible at any point.
• LOD techniques are critical
• Terrain meshes can be extremely dense.
• The U.S. Geological Survey data 30-arc-second
  resolution (roughly 1 kilometer at the equator)
  933 million points, 1.8 billion triangles over
  the entire planet.

28
Algorithms for Terrain(Top Down or Bottom Up)

• Top Down Subdivision or Refinement methods
• Begin with two or four triangles for the entire
  region.
• Progressively add new triangles until the desired
  resolution is achieved.
• Bottom Up Decimation or Simplification
• Begins with the highest-resolution mesh
• Iteratively removes vertices from the
  triangulation until the desired level of
  simplification is gained.

29
Algorithms for Terrain(Top Down or Bottom Up)

• Bottom-up approaches tend to be able to find the
  minimal number of triangles required for a given
  accuracy.
• However, they necessitate the entire model being
  available at the first step.
• Higher memory and computational demands.

30
Algorithms for Terrain(Top Down or Bottom Up)

• Bottom-up approaches are almost always used
  during the initial offline hierarchy
  construction.
• At run-time, a top-down approach might be
  favored.
• It offers support for view culling.

31
Algorithms for Terrain(Regular Grids and TINs)

• The use of regular grid height fields
• Regular (uniform) grids use an array of height
  values at regularly spaced x and y coordinates.
• Triangulated Irregular Networks (TINs)
• TINs allow variable spacing between vertices.

32
Algorithms for Terrain(Regular Grids and TINs)

• Advantages of TINs
• They can approximate a surface to a required
  accuracy with fewer polygons.
• Large flat regions are represented with a coarse
  sampling.
• Higher sampling is reserved for more bumpy
  regions.
• They offer great flexibility in the range and
  accuracy of features.
• Ridges, valleys, coastlines, caves, etc.
• Disadvantages of TINS
• They make implementing related functions (view
  culling, terrain following, collision detection
  and dynamic deformation) more complex.
• Due to the lack of a simple overarching spatial
  organization.
• The applicability of TINs to run-time
  view-dependent LOD is less efficient than regular
  gridded systems.

33
Algorithms for Terrain(Regular Grids and TINs)

• Disadvantages of Regular Grids
• They tend to be far less optimal than TINs.
• The same resolution is used across the entire
  terrain.
• Advantages of Regular Grids
• They are simple to store and manipulate.
• They are easily integrated with raster databases
  and file formats
• DEM, DTED, GeoTIFF, etc.
• They require less storage for the same number of
  points.
• An array of z values needs to be stored rather
  than full (x,y,z) coordinates.

34
Algorithms for Terrain(Regular Grids and TINs)

• For these reasons, many contemporary terrain LOD
  systems favor regular grids over TINs.

35
Algorithms for Terrain(Quadtrees and Bintrees)

• For multiresolution representation we use
  quadtrees or bintrees.
• Quadtree Structure
• A rectangular region is divided uniformly into
  four quadrants.
• Each of these quadrants can then be successively
  divided into four smaller regions.

36
Algorithms for Terrain(Quadtrees and Bintrees)

• A binary triangle tree structure works the same
  way as a quadtree.
• But it segments a triangle into two halves.
• The root triangles is normally defined to be a
  right-isosceles triangle.
• The subdivision is performed by splitting this
  along the edge formed between its apex vertex and
  the midpoint of its base edge.

37
Algorithms for Terrain(Quadtrees and Bintrees)

• Advantages of Bintrees
• They make it easy to avoid cracks and
  T-junctions.
• Exhibit the useful feature that triangles are
  never more than one resolution level away from
  their neighbors.

Subdivision progression example.
The root triangle is A
38
Algorithms for Terrain(Tears, Cracks and
T-Junctions)

• When adjacent triangles exist at different levels
  of detail,
• It is possible to introduce cracks along the
  edge.
• The higher LOD introduces an extra vertex that
  does not lie on the lower LOD edge.
• When rendered, these cracks can cause holes in
  the terrain, allowing the background to peak
  through.

39
Algorithms for Terrain(Tears, Cracks and
T-Junctions)

• When adjacent triangles exist at different levels
  of detail,
• Another undesirable artifact is the T-junction.
• It is caused when the vertex from a higher LOD
  triangle does not share a vertex in the adjacent
  lower LOD triangle.
• It can result in bleeding tears in the terrain
  and visible lighting and interpolation
  differences across the edges.

40
Algorithms for Terrain(Tears, Cracks and
T-Junctions)
41
Algorithms for Terrain(Tears, Cracks and
T-Junctions)

• How to deal with cracks
• The triangles around the crack are recursively
  split to produce a continuous surface.
• It is often used in bintree-based system.

42
Terrain Texturing

• After generating and rendering the geometry of
  the landscape, we still need to texture it.
• It means covering the triangles making up the
  landscape with images of the ground in that area.
• Method 1
• Just repeat a texture over the landscape.
• Results in visible tiling artifacts, and a quite
  boring landscape.

43
Terrain Texturing

• Method 2
• Use a number of different textures which match
  each other along some edges.
• We fill the plane with these textures, making
  sure adjacent texture edges match each other.
• Still somewhat boring landscape.
• Method 3
• To make the landscape more varying, we can
  texture different ecotopes with different
  textures.
• Mountain tops and hill slopes can be bare while
  valley floors can be more lush.
• We blend between different textures based on the
  strength of different ecotopes.

44
Terrain Texturing

• Texture Blending
• Instead of simply blending between different
  terrain textures, more natural edges can be
  achieved by using custom textures for the edges
  between textures.

45
Terrain Texturing

• Texture Blending
• The edge tiles can be drawn so that the
  underlying area is left transparent, allowing the
  topmost texture to be simply drawn on top of
  underlying textures.

46
Examples of Terrain Rendering Texturing