bul 2004 
and CSG 
' CSG to 
> 4 shows 
Rep. 
  
Rep 
isands of 
er, many 
cometric 
f typical 
lem is 
city 3d 
| in large 
) reduce 
D scene 
at high 
aging of 
r texture 
frustum, 
Culling , 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part BS. Istanbul 2004 
and LOD (level of detail ) technique in rendering of large area 
3D scene (Figure.5). [Woodl  1999][  Rossignac 
1993][Duchaineau 1997] 
  
   
  
    
         
  
Object Culling 
   
  
  
\ View axis 
Asa ef 
ack Face Culling. 
View Range ; | 
‘ Bx]! 
(IO 
/ / 
/ 1 
I 
I 
I 
    
Fig.5 3D Scene Accelerate Rendering strategy 
Using object culling, those sightless objects will be culling and 
only those visual objects can be rendering. For the object’s 
visibility, we can use triangle to describe the view range 
(Figure.5) and judge whether the object is in triangle. The 
object’s visibility can be defined as: 
: v>0 inside 
IIx » 1 s 
e v< 0 outside 
(z123537;2123) 
Where, (X, y) is the coordinate of object, (x,,7,) is 
triangle vertex coordinate. 
For those visual objects, we can give it a weight coefficient 
according to its weightiness and its position away from 
viewpoint to decide its detail. So we use a function 
D(O,d,m) to describe object detail. 
ne 0 d>d, 
"| D(O,d,m) d<d, 
Where, O is object id, d is distance which object away from 
viewpoint, m is object weight coefficient. D is object detail. 
According to D, we can use different level of detail (LOD) to 
describe and render the object. 
(2) 
3.2 LOD Terrain Rendering 
In Large 3D scene, we will spent a lot of resources on rendering 
terrain. The terrain data include digital elevation model (DEM) 
and texture. Currently, methods of LOD terrain rendering can be 
divided into two parts, GRID LOD and TIN LOD[Hamann 
1990][ Hoppe 1996][Hoppe 1998][ Duchaineau 1997]. the 
GRID terrain data are simplified in bintree or quadtree. 
In this paper, we divided the GRID terrain into some patches 
which its size must be power(2),for example ,32x32,64x64. we 
first judge whether the patch is visible, then use bintree to 
splitting the visible patch(Figure 6). When the terrain be 
rendering, the detail of level can be decided by the weight of 
bintree node calculated in advance and the distance from node 
io viewpoint. 
     
  
  
  
  
  
  
Fig6 bintree partition 
Binary Triangle Trees are used to store the splitting process. 
Binary Triangle Trees can be defined as: 
struct TriTreeNode 
{ TriTreeNode *LeftChild; 
TriTreeNode *RightChild; 
TriTreeNode *BaseNeighbor; 
TriTreeNode *LefiNeighbor; 
TriTreeNode *RightNeighbor; 
2. 
fa 
Figure.7 shows the topologic relationship of Binary Triangle 
Trees. 
  
  
  
  
Base Neighbor 
; Left 
Left MW Child 
Neighbor 
Righ 
Child 
Right 
Neighbor 
Fig.7 topologic relationship of binary triangle trees 
The following Figure.8 is LOD DEM patches render based on 
view range culling. 
  
Fig.8 DEM patches culling 
/S a birds-eve view of the LOD terrain 
pe 
Deos Ee ree An 
HN A e 
zs ES 
C ee 
de