Upload
shanghai-jiaotong-university
View
442
Download
0
Tags:
Embed Size (px)
Citation preview
Background Subtraction Based on Phase and Distance
Transform Under Sudden Illumination Change
Authors: Gengjian Xue, Jun Sun, Li Song
Reporter: Li Song
Shanghai Jiao Tong University
1 2012-09-27
Outline
Introduction
Background subtraction based on phase
feature and distance transform
Experimental results
Conclusion
2 2012-09-27
Introduction (1)
Moving object detection is an active
research subject in computer vision.
Foreground detection under sudden
illumination change is still a challenging
problem.
3 2012-09-27
Introduction (2)
Parametric density estimation technique
Gaussian Mixture Models (GMM), C.Stauffer and
Grimson, CVPR, 1999.
Nonparametric density estimation technique
Kernel Density Estimation (KDE), A. Elgammal,
etc., Pro.IEEE, 2002.
Region-based technique
Local Binary Pattern based (LBP), M.Heikkila and
M. Pietikainen, TPAMI, 2006
4 2012-09-27
Introduction (3)
Other methods
Wallflower system, K.Toyama, etc., ICCV, 1999
Pixel order information based, Binglong Xie, etc.,
Image Vision Computing, 2004
Pilet’s method, ECCV 2008.
---a statistical model combing two texture-based
features and the ratios of the current image to the
background image in each color channel
5 2012-09-27
Motivation
Phase is insensitive to illumination
change.
The basic framework of GMM method.
Blob aggregation using distance
transform.
6 2012-09-27
Diagram
7 2012-09-27
Background subtraction
based on phase feature
and distance transform
Phase feature extraction (1)
The Gabor wavelet coefficient:
:amplitude item
:phase item
Each pixel has wavelet coefficients
8 2012-09-27
))(exp()()( ,,, zizAzG vz
)(, zA v
)2,0[)(, z
maxmax
Phase feature extraction (2)
Phase extraction principle:
The larger amplitude of the coefficient is, the
more efficient the coefficient represents the pixel
information and the more discriminative of the
corresponding phase is.
Multiple phase extraction reasons:
rotation property
range of single phase is small
9 2012-09-27
Phase feature extraction (3)
The phase feature is defined:
: the phase feature
: the Gabor phase corresponding to the largest Gabor
amplitude of the pixel
L : the number of phase to be extracted
image subdivision
10 2012-09-27
L
i
i zzp1
)()(
)(zp
)(zi thi
NN
Background modeling
mixture of Gaussian distributions
Gaussian distribution represented by
, and , total weights
11 2012-09-27
thk
k2
k
K
i
i
1
1k
K
In the background modeling stage, each pixel is modeled
independently by a mixture of Gaussian distributions
where each Gaussian represents the phase distribution.
Model updating (1)
Set ,
Singular regions exists, matching condition
is in different expressions
if and
if and
others
12 2012-09-27
)6,0[)( zp3L
kkt
ktk
ktk
X
X
X
5.2
5.26
5.26 tX k6
k tX6
Model updating (2)
Condition satisfied
1. if and
2. if and
3. Others
13 2012-09-27
k tX6
222 )6()1(
)6(
ktkk
ktkk
X
X
tX k6
222 )6()1(
)6(
ktkk
ktkk
X
X
222 )()1(
)(
ktkk
ktkk
X
X
Model updating (3)
Weights update
is beyond
if
if
Condition not satisfied
New Gaussian replace , low
14 2012-09-27
inittX , init
kkk M )1(
k )6,0[
6
6
kk
kk
0
6
k
k
Foreground detection
Sorted in
Background distributions:
15 2012-09-27
)min(arg1
TBb
kk
Distance transform (1)
Initial detection result
16 2012-09-27
Distance transform (2)
Euclidean distance transform of pixel
17 2012-09-27
22
),()()(min),( yjxijid
Wyx
),( jia
Blob aggregation
Example results before and after blob aggregation
18 2012-09-27
Experiments (1)
Several sequences for testing.
GMM and LBP methods are employed for
comparison.
Visual and numerical evaluation are used
Precision and Recall as the metrics.
19 2012-09-27
100%FPTP
TPPrecision
100%
FNTP
TPRecall
Experiments (2)
Parameter selection
Gabor filter:
Phase extraction:
Gaussian model:
Segment value for DT:
20 2012-09-27
3,4 LN
4max 6max
7.0,3 bTK
2.1dT
Experiments (3)
21 2012-09-27
(a) (b) (c)
(d)
GMM
(e)
LBP
(f)
Our
Experiments (4)
Precision Recall
GMM 22.03 76.40
LBP 18.40 73.05
Our method 78.53 97.48
22 2012-09-27
Precision and recall evaluation
Experiments (5)
23 2012-09-27
(a) (b) (c)
(d)
GMM
(e)
LBP
(f)
Our
Experiments (6)
Precision Recall GMM 4.52 26.01 LBP 14.27 46.91
Our method 73.58 92.25
24 2012-09-27
Precision and recall evaluation
Conclusions
Pixel phase as feature.
Novel matching condition and updating
scheme.
Distance transform on the initial
detection results.
25 2012-09-27