MSThesisPresentation

Information about MSThesisPresentation

Published on February 7, 2008

Author: Berta

Source: authorstream.com

Content

Vehicle Recognition in Cluttered Environments:  Vehicle Recognition in Cluttered Environments Masters Thesis Defense By Gerald Dalley Signal Analysis and Machine Perception Laboratory The Ohio State University 05 June 2002 Overview:  Overview Problem Statement and Motivation Recognition Steps Range Image Generation Local Surface Estimation and Decimation Global Surface Reconstruction Surface Segmentation Graph Matching Conclusions and Future Work Questions Problem Statement and Motivation:  Problem Statement and Motivation Problem Recognize vehicles Military and civilian Forested environment Motivation Hostile forces tend to hide Camouflage and occlusion foil the human visual system Range Image Generation: Overview:  Range Image Generation: Overview Objects modeled Clutter models Camera flight paths (scenes) Noise generation Range Image Generation Local Surface Fitting Surface Reconstruction Surface Segmentation Graph Matching Range Image Generation: Objects Modeled:  Range Image Generation: Objects Modeled Range Image Generation: Clutter Models:  Range Image Generation: Clutter Models Visually realistic trees Look good, but… Poor occlusion Very long runtimes Range Image Generation: Camera Flight Paths (Scenes):  Range Image Generation: Camera Flight Paths (Scenes) circle flyby unoccluded Range Image Generation: Noise Generation:  Range Image Generation: Noise Generation Isotropic additive Gaussian noise Standard deviations of: 0mm 2mm 4mm 8mm 16mm 32mm Local Surface Estimation and Decimation: Overview:  Local Surface Estimation and Decimation: Overview Assumption: Vehicles are composed primarily of large, low-order, low-curvature surfaces. Constraint: 10 tank views  more than 220,000 range points (too many) Point Selection (Decimation) Principle Component Analysis Biquadratic Surface Fits Range Image Generation Surface Reconstruction Surface Segmentation Graph Matching Local Surface Estimation Local Surface Estimation and Decimation: Point Selection:  Local Surface Estimation and Decimation: Point Selection Fit errors away from corner Fit errors due to the corner Fit errors due to noise Fit errors due to the corner Method 1: Randomly select 1% (for example) of the original points Make local surface estimates based on selected points Problems: No noise σ = 0.3 Local Surface Estimation and Decimation: Point Selection (cont’d.):  Local Surface Estimation and Decimation: Point Selection (cont’d.) Method 2: In the region of interest… Collect range image points into cubic voxel bins (128x128x128mm) Discard bins that have: Too few points Points that do not represent biquadratic surfaces well Retain only the centroids of the bins and their surface fits Local Surface Estimation and Decimation: Principle Component Analysis:  Local Surface Estimation and Decimation: Principle Component Analysis u w Local Surface Estimation and Decimation: Biquadratic Surface Fits:  Local Surface Estimation and Decimation: Biquadratic Surface Fits pi Global Surface Reconstruction: Overview:  Global Surface Reconstruction: Overview Motivations Post-Processing Range Image Generation Local Surface Fitting Surface Reconstruction Surface Segmentation Graph Matching Global Surface Reconstruction: Motivations:  Global Surface Reconstruction: Motivations Easy, unambiguous nearest-neighbor identification Fast searches over small cardinality Makes rendering easier Avoids incorrect groupings of nearby surfaces Global Surface Reconstruction: Post-Processing:  Global Surface Reconstruction: Post-Processing Surface Segmentation: Overview:  Surface Segmentation: Overview Motivation: Correspondence is hard Some Techniques Not Used Spectral Clustering An overview Normalized cuts Our affinity measure Results Range Image Generation Local Surface Fitting Surface Reconstruction Surface Segmentation Graph Matching Surface Segmentation: Some Techniques Not Used:  Surface Segmentation: Some Techniques Not Used Robust Sequential Estimators (Mirza) Regions of Constant Curvature (Srikantiah) Surface Segmentation: Overview of Spectral Clustering:  Surface Segmentation: Overview of Spectral Clustering Two surface points have an affinity… Aij y1, where Ayi=li yi Surface Segmentation: Normalized Cuts:  Surface Segmentation: Normalized Cuts Surface Segmentation: Our Affinity Measure:  Surface Segmentation: Our Affinity Measure Surface Segmentation: Unoccluded Results:  Surface Segmentation: Unoccluded Results Surface Segmentation: Which Objects Are These?:  Surface Segmentation: Which Objects Are These? Graph Matching: Overview:  Graph Matching: Overview Match tree example Error measures Entropy Results What caused problems? Range Image Generation Local Surface Fitting Surface Reconstruction Surface Segmentation Graph Matching Graph Matching: Match Tree Example:  Graph Matching: Match Tree Example Graph Matching: Error Measures:  Graph Matching: Error Measures Unary Error Area, Elongation, Thickness Orientation Error How poorly pairs of normals match up Centroid Distance Error How poorly pairs of centroids match up Cumulative Area Error What percentage of the model area is not matched up Graph Matching: Entropy:  Graph Matching: Entropy Graph Matching: Results: earthmover:  Graph Matching: Results: earthmover Graph Matching: Results: obj1:  Graph Matching: Results: obj1 Graph Matching: Results: sedan:  Graph Matching: Results: sedan Graph Matching: Results: semi:  Graph Matching: Results: semi Graph Matching: Results: tank:  Graph Matching: Results: tank Graph Matching: What Caused Problems?:  Graph Matching: What Caused Problems? 19 total incorrect recognition results 12: over-segmentation 10: area errors (including non-existent segments) Graph Matching: What Caused Problems? (cont’d.):  Graph Matching: What Caused Problems? (cont’d.) 4: mis-aligned segmentation Conclusions:  Conclusions System features Modular design Handles pessimistic levels of clutter 100% recognition on earthmover and sedan Reliable segmentation is important when doing graph matching Future Work:  Future Work Articulation Larger modelbase Iterative recognition Alternative segmentation methods Other affinity matrix normalizations Tensor voting Enhanced version of Srikantiah’s algorithm Verification Alternative recognizers (e.g. SAI) E3D! (hopefully, for the remaining SAMPL crowd) Your Questions...:  Your Questions... EXTRA SLIDES:  EXTRA SLIDES Range Image Generation: Ray Tracing:  Range Image Generation: Ray Tracing xs Global Surface Reconstruction: Preliminaries: Voronoi Diagrams:  Global Surface Reconstruction: Preliminaries: Voronoi Diagrams Voronoi cell = locus of points closer to a given sample point than any other point Global Surface Reconstruction: Preliminaries: Medial Axis:  Global Surface Reconstruction: Preliminaries: Medial Axis Medial axis = locus of points equidistant from at least two surface points (considering the original surface) Global Surface Reconstruction: Preliminaries: ε-sampling:  Global Surface Reconstruction: Preliminaries: ε-sampling ε-sampling = Samples are at most ε times the distance to the medial axis Global Surface Reconstruction: Cocone:  Global Surface Reconstruction: Cocone p+ ≡ pole of p = point in the Voronoi cell farthest from p ε < 0.06 → the vector from p to p+ is within π/8 of the true surface normal The surface is nearly flat within the cell Voronoi cell of p p+ p Surface Segmentation: Normalized Cuts:  Surface Segmentation: Normalized Cuts Surface Segmentation: Probabilistic Affinity Framework:  Surface Segmentation: Probabilistic Affinity Framework Surface Segmentation: Probabilistic Position Affinity:  Surface Segmentation: Probabilistic Position Affinity Surface Segmentation: Probabilistic Position Affinity:  Surface Segmentation: Probabilistic Position Affinity Surface Segmentation: Probabilistic Normal Affinity:  Surface Segmentation: Probabilistic Normal Affinity Surface Segmentation: Probabilistic Normal Affinity:  Surface Segmentation: Probabilistic Normal Affinity Graph Matching: Error Measures:  Graph Matching: Error Measures Graph Matching: Error Measures (cont’d.):  Graph Matching: Error Measures (cont’d.) Graph Matching: Error Measures (cont’d.):  Graph Matching: Error Measures (cont’d.)

Related presentations


Other presentations created by Berta

CII PRESENTATION
09. 01. 2008
0 views

CII PRESENTATION

John Wilkes Booth
10. 01. 2008
0 views

John Wilkes Booth

Abdala Amphetamines in Russia
10. 01. 2008
0 views

Abdala Amphetamines in Russia

terzalezione scritturanarrativa
12. 01. 2008
0 views

terzalezione scritturanarrativa

Colonialism
15. 01. 2008
0 views

Colonialism

Evolution for Beginners
17. 01. 2008
0 views

Evolution for Beginners

The Analysis Gatsby
04. 02. 2008
0 views

The Analysis Gatsby

OCTranspo
05. 02. 2008
0 views

OCTranspo

mhd day2005
16. 01. 2008
0 views

mhd day2005

nut info labels etiquettes e
07. 02. 2008
0 views

nut info labels etiquettes e

american revolution
12. 02. 2008
0 views

american revolution

moorthy ms
17. 01. 2008
0 views

moorthy ms

leadinglearning
20. 02. 2008
0 views

leadinglearning

izellerbauhauscli2005
26. 02. 2008
0 views

izellerbauhauscli2005

PersuasiveComp
28. 02. 2008
0 views

PersuasiveComp

ADA overview
07. 03. 2008
0 views

ADA overview

wipo smes uln 07 www 89154
14. 03. 2008
0 views

wipo smes uln 07 www 89154

latair
22. 01. 2008
0 views

latair

Presentation UWIHARP Final
19. 01. 2008
0 views

Presentation UWIHARP Final

Stuart Anderson
03. 04. 2008
0 views

Stuart Anderson

venkatesh
08. 04. 2008
0 views

venkatesh

identifiers
09. 01. 2008
0 views

identifiers

2007112679352641
14. 04. 2008
0 views

2007112679352641

Marketing of Evil book review
16. 04. 2008
0 views

Marketing of Evil book review

21431
17. 04. 2008
0 views

21431

IR3001 Middle East security
23. 04. 2008
0 views

IR3001 Middle East security

DSRDSept19
24. 04. 2008
0 views

DSRDSept19

BasevsDIStudio
05. 02. 2008
0 views

BasevsDIStudio

TNG
02. 05. 2008
0 views

TNG

Gavin Mooney DRS Sept 06
14. 02. 2008
0 views

Gavin Mooney DRS Sept 06

STD11
02. 05. 2008
0 views

STD11

CLOCgonzalez
15. 01. 2008
0 views

CLOCgonzalez

9 10 07Watson Presentation2
30. 01. 2008
0 views

9 10 07Watson Presentation2

f3
13. 01. 2008
0 views

f3

gep2004slideshow
24. 01. 2008
0 views

gep2004slideshow

LuisLobopresentation
17. 01. 2008
0 views

LuisLobopresentation

adcas
16. 01. 2008
0 views

adcas

Weber Paul
14. 01. 2008
0 views

Weber Paul

BoraBora
29. 01. 2008
0 views

BoraBora

swga 004
05. 03. 2008
0 views

swga 004