An Efficient Algorithm for Feature-based 3D Point Cloud - - PowerPoint PPT Presentation

an efficient algorithm for feature based 3d
SMART_READER_LITE
LIVE PREVIEW

An Efficient Algorithm for Feature-based 3D Point Cloud - - PowerPoint PPT Presentation

Zili Yi Co-authors: Yang Li, Minglun Gong Memorial University of Newfoundland 2016-12-12 An Efficient Algorithm for Feature-based 3D Point Cloud Correspondence Search Outline Introduction Background Method Results


slide-1
SLIDE 1

An Efficient Algorithm for Feature-based 3D Point Cloud Correspondence Search

Zili Yi Co-authors: Yang Li, Minglun Gong Memorial University of Newfoundland 2016-12-12

slide-2
SLIDE 2

Outline

 Introduction  Background  Method  Results  Conclusion

slide-3
SLIDE 3

Introduction – Point Cloud

 3D point cloud

 A versatile representation of 3D shapes

 Sources

 laser scanners  estimated from stereo matching  sampled from 3D surface models.

slide-4
SLIDE 4

Introduction - 3D Point Cloud Correspondence

 3D Point Cloud Correspondence

 to find the matching between

the two sets of points

 Application

 building statistical shape models  smoothly interpolating key

Frames in cartoon animations

 morphing between shapes of disparate

  • bjects

 recognizing/classifying 3D objects 3D Point Cloud Correspondence

slide-5
SLIDE 5

Introduction – previous methods

 Methods for Point Cloud Correspondence

 Rigid (transformation to correspondence)

 ICP

, BMP

 Non-rigid (Pointwise/articulated correspondence)

 Feature-based  Non-rigid BMP/ICP

 Feature-based Correspondence

 Matching between features rather than raw points

 Limitation of traditional feature-based algorithm

 Computation of features at multiple levels  Inefficient when introducing smoothness term

slide-6
SLIDE 6

Related topics

 Nearest Neighbor Search

 Space partitioning trees: K-d tree,

VP-tree

 PatchMatch (for 2D image correspondence search)

 Combine randomized search and belief propagation

 3D Point Cloud Features

 Unique Shape Context  Point Feature  Point Feature Histogram

slide-7
SLIDE 7

Related topics

 Swarm Intelligence

 Artificial Bee Colony (ABC)  Search optimal solutions

 Three types of bees in a Colony

 Scout

 Search food sources randomly

 Employed bee

 Search food sources among neighbors

 Onlooker

 Search food sources from other colonies

slide-8
SLIDE 8

Method Framework

Objective Function Optimization Correspondence Result Preparation 3D Feature Extraction Colorization K-Neighbor Caching Random Initialization Source Target ABC-based Search Converge Source Target

slide-9
SLIDE 9

Method

 Objective

 S: Source, T: Target  P ∈ 𝑇, M(P) is the matching point of P 

α is the balance coefficient

 Geometric term  Smooth term

slide-10
SLIDE 10

Effect of Smoothness Term

 Force neighbors corresponding to neighbors

slide-11
SLIDE 11

Optimization Process

 Searching scheme (iteratively and pointwisely)

slide-12
SLIDE 12

Optimization Process

slide-13
SLIDE 13

Optimization Technique

 Variable harmonic vs fixed harmonic during optimization

Source Target Source α=0 α varying 0-0.95 α=0.95 (mesh) (mesh) (point cloud) result I result II result III

slide-14
SLIDE 14

Qualitative Comparison

slide-15
SLIDE 15

Efficiency

An order of magnitude faster than brute-force search

slide-16
SLIDE 16

Conclusion

 Strength

 Efficient  Accurate  Noise-robust

 Limitation

 Poor matching with giant shape difference Correspondence results on the tiger-horse dataset under different noise levels. In each test, both the target (left) and the source (right) point clouds are corrupted using Gaussian noise with standard

variance of σ, where D is the diagonal length of the input point cloud.

slide-17
SLIDE 17

Conclusion

 Strength

 Efficient  Accurate  Noise-robust

 Limitation

 Poor matching between shapes with giant difference Correspondence results on the tiger-horse dataset under different noise levels. In each test, both the target (left) and the source (right) point clouds are corrupted using Gaussian noise with standard

variance of σ, where D is the diagonal length of the input point cloud.

slide-18
SLIDE 18

Thank You!