Motion Segmentation - PowerPoint PPT Presentation

1 / 23
About This Presentation
Title:

Motion Segmentation

Description:

IEEE Transactions on Pattern Analysis and Machine ... Deteriorating as n increases. C is sensitive to outliers. C. A. G. D. C. G [2] LSA. clustering ... – PowerPoint PPT presentation

Number of Views:253
Avg rating:3.0/5.0
Slides: 24
Provided by: mathZ
Category:

less

Transcript and Presenter's Notes

Title: Motion Segmentation


1
Motion Segmentation
  • CAGDCG Seminar
  • Wanqiang Shen
  • 2008-04-09

2
Application
3
Motion analysis
Motion segmentation
4
Problem
  • Accurate
  • Robust
  • Fast

How much
What
How
5
Traditional model
  • A rigid-body motion
  • Multiple rigid-body motions

6
Paper
  • 1 R. Vidal, Y. Ma, and S. Sastry. Generalized
    Principal Component Analysis (GPCA). IEEE
    Transactions on Pattern Analysis and Machine
    Intelligence, 27(12)115, 2005.
  • 2 J. Yan and M. Pollefeys. A general framework
    for motion segmentation Independent,
    articulated, rigid, non-rigid, degenerate and
    non-degenerate. In European Conference on
    Computer Vision, pages 94106, 2006.
  • 3 R. Tron and R. Vidal A Benchmark for the
    Comparison of 3-D Motion Segmentation Algorithms.
    IEEE International Conference on Computer Vision
    and Pattern Recognition, 2007.

7
1 GPCA
Model
8
1 Model
9
1 Estimating n
10
1 Estimating subspaces
  • calculating normalized C
  • Factorization
  • Solving for the last 2 entries of each bi
  • Solving for the first K-2 entries of each bi

11
1 Optimizing clustering
12
1 example
13
1 Remarks
  • Advantages
  • Algebraic algorithm
  • Dealing with both independent and dependent
    motions
  • disadvantages
  • Deteriorating as n increases
  • C is sensitive to outliers

14
2 LSA
15
2 Projection
16
2 Local subspace estimation
Affinity matrix
17
2 Clustering
  • Estimation N
  • While Numofclusterslt N
  • Compute affinity matrix for each clusters
  • Divide each cluster into two clusters
  • Evaluate the best subdivision

18
2 examples
19
2 Remarks
  • Advantages
  • Outliers are likely to be rejected
  • Need less point trajectories
  • disadvantages
  • Neighbors of a point belong to different subspace
  • The select neighbors may not span the underlying
    subspace

20
3 test samples
21
3 Benchmark
22
3 comparing data
  • Two groups

accuracy GPCA LSA
Check. 6.09 5.71
Traffic 1. 41 3.75
Articul. 2.88 4.38
All 4.59 5.09
time GPCA LSA
Check. 353ms 7.762s
Traffic 288ms 6.787s
Articul. 224ms 4.002s
All 324ms 7.165s
  • Three groups

accuracy GPCA LSA
Check. 31.95 18.09
Traffic 19.83 26.05
Articul. 16.85 15.18
All 28.66 19.51
time GPCA LSA
Check. 842ms 17.314s
Traffic 529ms 12.746s
Articul. 125ms 1.288s
All 738ms 15.485s
23
Thank you!
Write a Comment
User Comments (0)
About PowerShow.com