Observer Relative Data Extraction Linas Bukauskas 3DVDM group Aalborg University, Denmark 2001.

Observer Relative Data Extraction

Linas Bukauskas

3DVDM group

Aalborg University, Denmark

2001

2001 N/X VMMD Workshop 2001 2

Content• Motivation• Observer Relative Data Extraction

– Visibility Range– Tree Structure– Visibility cases

• Experimental Results• Related Work• Conclusions• Future Work

2001 N/X VMMD Workshop 2001 3

Motivation

• Unbounded Universe of objects– CAVE® and Panorama creates fully immersed

environment

• All objects are not visible at once– Catalog of stars 50GB

(donor: Jim Gray @ Microsoft Research)

2001 N/X VMMD Workshop 2001 4

Motivation (cont`d)

• Visualization system can not handle all objects in the Universe– Rendering of the world is time consuming

• Observer is moving through the Universe– Arriving objects appear, leaving - disappear

2001 N/X VMMD Workshop 2001 5

Example of Moving Observer in 3D

2001 N/X VMMD Workshop 2001 7

ORDE Queries

• Objects that are visible• Objects of specific visibility level

• Objects that will become (in) visible• Objects that might be visible soon

• Objects that might be visible moving along the path

2001 N/X VMMD Workshop 2001 8

Distance Based Organization

• Create tree structure to access data

• Use distance based organization– Visibility Factor a parameter in a node

The tree will order objects according the visibility factor

– Second storage access structureB-Tree like structure

2001 N/X VMMD Workshop 2001 9

Distant Based Organization

• Visibility Factor

• Visible Objects

2001 N/X VMMD Workshop 2001 10

Distant Based Organization Fails

• Objects far away can be visible (if large)

• Near objects can be invisible (if small)

1

2

2001 N/X VMMD Workshop 2001 11

Observer Relative Data Extraction

• Requirements– Static Visibility Factor– Cluster/partition the space– Hierarchical structure– Second storage structure

2001 N/X VMMD Workshop 2001 12

Visibility Range (cont´d)

• Definition: Let Oi be an object. The visibility range associated with the object, VRi(Oi) is:

• VR is a Minimal Bounding Square (MBS)

• Brightness and color can be incorporated

2001 N/X VMMD Workshop 2001 13

Visibility Range

• Overlapping object visibility ranges.

MBS

VR

2001 N/X VMMD Workshop 2001 14

The Tree Structure

• Hierarchical structure of MBRs and MBSs

1

23

4

7

65

1 2 3 4 5 6 7

2001 N/X VMMD Workshop 2001 15

The Tree Structure

• Querying: Overlaps

1

23

4

7

65

1 2 3 4 5 6 7

2001 N/X VMMD Workshop 2001 16

The Tree Structure (Cont´d)

• Two types of nodes:– MBRs internal– MBSs leaf nodes

• Pack more objects into leaf 1KB nodes2D 3D

Internal 256 170Leafnode 341 256

2001 N/X VMMD Workshop 2001 17

Three Cases of Queries

• Perfect– Visibility Ranges are as is

• Conservative– Visibility Ranges are enlarged

• Optimistic– Visibility Ranges are reduced

2001 N/X VMMD Workshop 2001 18

Perfect Case

• Point query – Observer position as input

– Extracts only Visible Objects

• Window Query– Region of movement

– Extracts now Visible Objects +Objects visible soon

Scale factor: r = 1

2001 N/X VMMD Workshop 2001 19

Conservative Case

• Point query – Observer position as input

– Surplus Visible Objects

– does not extract exactly Visible Objects

• Window Query– More surplus Visible Objects

Scale factor: r > 1

2001 N/X VMMD Workshop 2001 20

Optimistic Case

• Point query – Observer position as input

– Very Visible Objects

• Window Query– Region as input

– Ensure Visible Object extraction, surplus invisible.

Scale factor: 0 < r < 1

2001 N/X VMMD Workshop 2001 21

Three Cases of Queries

• Perfect – Finds exactly visible objects for the observer

• Conservative– Finds visible objects with a buffer for the

observer to move

• Optimistic– Optimistically extracts visible objects, with a

surplus amount of invisible data.

2001 N/X VMMD Workshop 2001 22

Experiments

• R-Tree vs. VR-Tree– Universe 100x100 units– Varying size of data set 250.000 - 1 mio. – Largest VR span 1% and 10% of the Universe– Page size 1 KB– Implemented on GIST

2001 N/X VMMD Workshop 2001 23

R vs. VR –Tree

0

20

40

60

80

100

120

140

0.25 0.5 0.75 1

# of objects in mln

I/O

per

ob

ject

2.3

2.4

2.5

2.62.7

2.8

2.9

3

0.25 0.5 0.75 1

# of objects in mln

10 % of universe1% of Universe

VR

R

2001 N/X VMMD Workshop 2001 24

Supernovas• Supernovas has impact in Optimistic case

– Perfect & Conservative vs. Optimistic

0

10000

20000

30000

40000

50000

0.25 0.5 0.75 1

# of objects in mln

IO

01000020000300004000050000600007000080000

0.25 0.5 0.75 1

# of objects in mln

2001 N/X VMMD Workshop 2001 25

Related Work

• R-Tree (A. Guttman. R-Trees: A Dynamic Index Structure for Spatial

Searching.1984)

– X-Tree (S. Berchtold, D. A. Keim, and H.-P. Kriegel. The X-tree : An Index Structure for

High-Dimensional data, 1996.)

– SS-Tree (D. A. White and R. Jain. Similarity Indexing with the SS-tree. 1996)

– SR-Tree (N. Katayama and S. Satoh. The SR-tree: An Index Structure for High-Dimensional Nearest Neighbor Queries.1997)

– TPR-Tree (S. Saltenis, C. S. Jensen, S. T. Leutenegger, and M. A. Lopez. Indexing the Positions of Continuously Moving Objects, 2000)

2001 N/X VMMD Workshop 2001 26

Related Work (cont’d)

• Space partitioning– Kd-Tree, Quad/Oct-Trees – kdB-Tree (J. T. Robinson. The K-D-B-Tree: A Search Structure For Large

Multidimensional Dynamic Indexes.1981)

– LSDh Tree (A. Henrich. The LSD h -Tree: An Access Structure for Feature Vectors. 1998)

2001 N/X VMMD Workshop 2001 27

Conclusions

• Work in progress– Observer position dependant queries– Visibility Ranges– Three special cases of queries

• Perfect, Conservative, Optimistic

– Empirical evaluation

2001 N/X VMMD Workshop 2001 28

Future Work

• Evaluate tree in a higher dimensions– Does it make sense in Virtual Reality setting?

• Incremental data extraction when moving– Incoming and leaving objects

• Retrieve data that will be visible along the path– Given a path points optimize data extraction

• Validate results with cases from the real life

2001 N/X VMMD Workshop 2001 29

Acknowledgment

• Michael Böhlen

• 3DVDM project members

2001 N/X VMMD Workshop 2001 30

Questions?