approximating terrain with over-determined laplacian pdes

1
Zhongyi Xie, Marcus A. Andrade, W. Randolph Franklin, Barbara Cutler, Metin Inanc, Daniel M. Tracy and Jonathan Muckell Rensselaer Polytechnic Institute, Troy, NY Approximating Terrain with Over-Determined Laplacian PDE s Abstract: We extend Laplacian PDE by adding a n ew equation to form an over-determine d system (ODETLAP) which can be used to approximate the whole terrain from a few isolated points. We compress th e terrain by selecting a few points w hich could later be lossily ‘decompre ssed’ using ODETLAP. Points selection algorithms include TIN, Visibility te st, Level Set Components and Regular Selection. ODETLAP: Two sets of equations make the system ove r-determined: Laplacian Equation: every non-border point is the average of its neighbors: 4z ij = z i-1,j + z i+1,j + z i,j-1 + z i,j+1 New equation: Some points are already know n: z ij = h ij Use a smoothness parameter R to interpolat e the two, R reflect the relative import ance of accuracy vs. smoothness. Adds capabilities to the classical system • Local maxima inference • Inconsistent data conflation ODETLAP Point Selection: 1. Incremental TIN to find most important points, then greedy insertion of worst points (Allows progressive transmission) 2. Regular grid of points (more points that compress better) (More compact) 3. Visibility Index to find points that represent the structure of the terrain 3. Accuracy vs. Size 1. Compression Results (TIN + Greedy ODETLAP) Encoding ODETLAP’s output: Code (x, y) separately from z: •Run-length encode the bitmap: •Delta code {z}, then use bzip2. •Approach information theoretic within 20% ODETLAP ODETLAP RMS ≤ RMS ≤ Max_RMS? Max_RMS? Original Original terrain terrain representati representati on on Initial Initial points points selection selection Set of Set of importan importan t points t points Refined points Refined points selection selection Augmented Augmented important point important point set set No No Yes Yes Exi Exi t t Reconstruct Reconstruct ed ed surface surface Data Size, bytes Compressio n ratio RMS Elev Error, m RMS Slope Error, deg hill1 1880 170:1 2.83 3.53 hill2 1962 163:1 4.06 8.06 hill3 1739 184:1 1.66 1.65 mtn1 1979 162:1 3.77 14.0 mtn2 2006 160:1 4.31 14.1 mtn3 2004 160:1 4.58 13.3 2. TIN + Greedy Elevation Comparison: Mtn2 Dataset (Compressed Size: 7641 bytes ) Original Original Terrain Terrain Reconstructed Terrain Reconstructed Terrain Points Selected Points Selected using Visibility using Visibility Index Index

Upload: nikkos

Post on 26-Jan-2016

32 views

Category:

Documents


3 download

DESCRIPTION

Approximating Terrain with Over-Determined Laplacian PDEs. Abstract : - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Approximating Terrain with Over-Determined Laplacian PDEs

Zhongyi Xie, Marcus A. Andrade, W. Randolph Franklin, Barbara Cutler, Metin Inanc, Daniel M. Tracy and Jonathan Muckell

Rensselaer Polytechnic Institute, Troy, NY

Approximating Terrain with Over-Determined Laplacian PDEs

Abstract: We extend Laplacian PDE by adding a new equation to form an over-determined system (ODETLAP) which can be used to approximate the whole terrain from a few isolated points. We compress the terrain by selecting a few points which could later be lossily ‘decompressed’ using ODETLAP. Points selection algorithms include TIN, Visibility test, Level Set Components and Regular Selection.

ODETLAP:

Two sets of equations make the system over-determined:

Laplacian Equation: every non-border point is the average of its neighbors:

4zij = zi-1,j+ zi+1,j + zi,j-1 + zi,j+1

New equation: Some points are already known:

zij = hij

Use a smoothness parameter R to interpolate the two, R reflect the relative importance of accuracy vs. smoothness.

Adds capabilities to the classical system

• Local maxima inference

• Inconsistent data conflation

ODETLAP Point Selection:

1. Incremental TIN to find most important points, then greedy insertion of worst points (Allows progressive transmission)

2. Regular grid of points (more points that compress better) (More compact)

3. Visibility Index to find points that represent the structure of the terrain

3. Accuracy vs. Size

1. Compression Results (TIN + Greedy ODETLAP)

Encoding ODETLAP’s output:

Code (x, y) separately from z:

• Run-length encode the bitmap:

• Delta code {z}, then use bzip2.

• Approach information theoretic within 20%

ODETLAPODETLAPRMS ≤RMS ≤Max_RMS?Max_RMS?

Original Original terrain terrain

representatirepresentationon

Initial Initial points points

selectionselection

Set of Set of importanimportant pointst points

Refined points Refined points selectionselection

Augmented Augmented important point important point

setset

NoNoYesYes

ExitExit

ReconstructReconstructed ed

surfacesurface

DataSize, bytes

Compression ratio

RMS Elev Error, m

RMS Slope Error, deg

hill1 1880 170:1 2.83 3.53

hill2 1962 163:1 4.06 8.06

hill3 1739 184:1 1.66 1.65

mtn1 1979 162:1 3.77 14.0

mtn2 2006 160:1 4.31 14.1

mtn3 2004 160:1 4.58 13.3

2. TIN + Greedy Elevation Comparison: Mtn2 Dataset (Compressed Size: 7641 bytes )

Original TerrainOriginal Terrain

Reconstructed TerrainReconstructed Terrain

Points Selected using Points Selected using Visibility IndexVisibility Index