Transcript 下載/瀏覽

Detecting Moving Objects,
Ghosts, and Shadows in
Video Streams
Rita Cucchiara, Costantino Grana, Massimo Piccardi and Andrea Prati
Adviser : Chih-Hung Lin
Speaker : Kuan-Ju Chen
Date
: 2009/02/19
1
Author
 Author
 R. Cucchiara, C. Grana, and A. Prati are with the
Dipartimento di Ingegneria dell’Informazione,
Universita` di Modena e Reggio Emilia, Via Vignolese,
905/b, 41100 Modena, Italy.
 . M. Piccardi is with Department of Computer Systems,
Faculty of IT, University of Technology, Sydney,
Broadway NSW 2007, Australia.
 Accept
 IEEE TRANSACTIONS ON PATTERN ANALYSIS
AND MACHINE INTELLIGENCE, VOL. 25, NO. 10,
OCTOBER 2003
 revised 23 Sept. 2002; accepted 25 Feb.2003
2
Contents
1
INTRODUCTION
2
DETECTING MOVING OBJECTS
3
SHADOW DETECTION
4
RESULTS
5
CONCLUSIONS
3
INTRODUCTION
Detection of moving objects in video
streams is the first relevant step of
information extraction in many
computer vision applications
 Video surveillance
 People tracking
4
INTRODUCTION
In this work, the models of the target
objects and their motion are unknown
the most widely adopted approach for
moving object detection with fixed
camera is based on background
subtraction
5
INTRODUCTION
Background Subtraction
 Two problems
• Accurately
• Reactivity
If the background model is neither
accurate nor reactive
 Ghosts
 Shadows
6
INTRODUCTION
Solution:
 Statistical combination
• Combination frames to compute the background
model
 Combine
• Combine the current frame and previous models
with recursive filtering to update the background
model
7
INTRODUCTION
 In this paper, we propose a novel simple method
that exploits all these features, combining them
so as to efficiently provide detection of moving
objects, ghosts, and shadows.
 The main contribution
 the integration of knowledge of detected objects,
shadows, and ghosts in the segmentation
process
 enhance object segmentation and background
update.
8
DETECTING MOVING OBJECTS
The aim
 Detect real moving objects.
 Avoiding detection of transient spurious
objects
Sakbot (Statistical And KnowledgeBased ObjecT detection)
 Statistics and knowledge of the segmented
objects to improve both background modeling
and moving object detection
9
DETECTING MOVING OBJECTS
 MVO (Moving visual object)
 set of connected points belonging to object
characterized by nonnull motion
 Ghost
 a set of connected points detected as in motion by
means of background subtraction, but not
corresponding to any real moving object
 Shadow
 a set of connected background points modified by a
shadow cast over them by a moving object
10
DETECTING MOVING OBJECTS
Original
Image
Background
Foreground
Shadow
Object
MVO
Shadow
Ghost
Shadow
MVO
Ghost
Taxonomy
11
DETECTING MOVING OBJECTS
Frame t
background model
Foreground
12
DETECTING MOVING OBJECTS
Get a statistical information
Follow set of
S
t  t
the statistical background model Bs  p 
Using the median function get a statistical information: Bkt  t  p 
If p belong to MVO , p=background model
If p dosen`t belong to MVO, p= statistical background model
13
DETECTING MOVING OBJECTS
t  t
used to update the background model to update background model B  p 
If p isn`t belong to known object , p= Bs  p 
If p is belong to known object , p= Bkt  t  p 
t  t
update background model
14
SHADOW DETECTION
Mean the process of classification of
foreground pixels as “shadow points”
based on their appearance
15
SHADOW DETECTION
Foreground
Ayalyze Hue-Saturation-Value(HSV) color space
Following three condition to mask shadow
identifying as shadows those points
define a darkening effect of shadows

I t  p .V
t
t



 p .S   s  D H   H ;  0,1,   0,1
1
if





I
p
.
S

B
t
t
shadow mask:SP  p 
B  p .V
0
otherwise
average image luminance
16
RESULTS
The reactivity of the background model.
17
RESULTS
Segmentation is provided via background subtraction including shadow detection.
(a) false positives and (b) false negatives.
18
CONCLUSIONS
 This paper has presented Sakbot, a system
for moving object detection in image
sequences
 This system has the unique characteristic of explicitly
addressing various troublesome situations such as
cast shadows and ghosts.
 Cast shadows are detected and removed from the
background update function
 Ghosts are also explicitly modeled and detected so as
to avoid a further cause of undesired background
modification
 The method is highly computationally costeffective since it is not severe in
computational time
19
Add your company slogan
20
DETECTING MOVING OBJECTS
MVOs are validated by applying a set of
rules on area, saliency, and motion as
follows:
 The MVO blob must be large enough
 The MVO blob must be a “salient” foreground
blob
 The MVO blob must have nonnegligible
motion
21
DETECTING MOVING OBJECTS
• Validation rules
22

Follow set of S
S  I  p , I
t
t - t
 p ,..., I
t - nt
 p  b B  p 
t
the statistical background model:
Bst  t  p   arg i 1,..., k min
 Distance x , x 
k
i
j 1
j
xi , x j  S
knowledge-based background model:

 

   
t
t
t



B
p
if
p

MVO

MVO
SH
Bkt  t  p    t  t
t
t


B
p
if
p

G

G
SH
 s
update the background model:
t  t

 p
B
t  t
s
B  p    t  t
 Bk  p 
if O  KOt : p  O
otherwise
23