Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more
Download
Standard view
Full view
of .
Save to My Library
Look up keyword
Like this
9Activity
0 of .
Results for:
No results containing your search query
P. 1
Adaptive Background Estimation and object detection applying in Automated visual surveillance

Adaptive Background Estimation and object detection applying in Automated visual surveillance

Ratings: (0)|Views: 542 |Likes:
Published by ijcsis
Automated visual surveillance is currently a hot topic in computer vision research. A common approach is to perform background subtraction, which identifies moving objects from the sequence of video frames that differs significantly from a background model. As a basic, the background image must be a representation of the scene with no moving objects and must be kept regularly updated. There are many challenges in developing a good background subtraction algorithm. We have proposed a methodology to perform background subtraction from moving vehicles in traffic video sequences that combines statistical assumptions of moving objects using the previous frames. It is necessary to update the background image frequently in order to guarantee reliability of the motion detection. For that, a binary moving objects hypothesis mask is constructed to classify any group of lattices as being from a moving object based on the optimal threshold. Then, the new incoming information is integrated into the current background image using a Kalman filter. In order to improve the performance, it is necessary to perform post- processing. It has been accomplished by shadow and noise removal algorithms operating at the lattice which identifies object-level elements. The results of post-processing can be used to detect object more efficiently. Experimental results and comparisons using real data demonstrate the superiority of the proposed approach which has achieved an average accuracy of 92% on completely novel test images.
Automated visual surveillance is currently a hot topic in computer vision research. A common approach is to perform background subtraction, which identifies moving objects from the sequence of video frames that differs significantly from a background model. As a basic, the background image must be a representation of the scene with no moving objects and must be kept regularly updated. There are many challenges in developing a good background subtraction algorithm. We have proposed a methodology to perform background subtraction from moving vehicles in traffic video sequences that combines statistical assumptions of moving objects using the previous frames. It is necessary to update the background image frequently in order to guarantee reliability of the motion detection. For that, a binary moving objects hypothesis mask is constructed to classify any group of lattices as being from a moving object based on the optimal threshold. Then, the new incoming information is integrated into the current background image using a Kalman filter. In order to improve the performance, it is necessary to perform post- processing. It has been accomplished by shadow and noise removal algorithms operating at the lattice which identifies object-level elements. The results of post-processing can be used to detect object more efficiently. Experimental results and comparisons using real data demonstrate the superiority of the proposed approach which has achieved an average accuracy of 92% on completely novel test images.

More info:

Published by: ijcsis on Aug 13, 2010
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

05/27/2013

pdf

text

original

 
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 8, No. 4, July 2010
Adaptive Background Estimation and object detectionapplying in Automated visual surveillance
M.Sankari
Department of Computer Applications,Nehru Institute of Engineering and Technology,Coimbatore, INDIA.sankarim2@gmail.com 
C. Meena
Head, Computer centre,Avinashilingam University,Coimbatore, INDIA.cccmeena@gmail.com 
 Abstract
 — 
Automated visual surveillance is currently a hot topicin computer vision research. A common approach is to performbackground subtraction, which identifies moving objects fromthe sequence of video frames that differs significantly from abackground model. A
s
a basic, the background image must be arepresentation of the scene with no moving objects and must bekept regularly updated. There are many challenges in developinga good background subtraction algorithm. We have proposed amethodology to perform background subtraction from movingvehicles in traffic video sequences that combines statisticalassumptions of moving objects using the previous frames. It isnecessary to update the background image frequently in order toguarantee reliability of the motion detection. For that, a binarymoving objects hypothesis mask is constructed to classify anygroup of lattices as being from a moving object based on theoptimal threshold. Then, the new incoming information isintegrated into the current background image using a Kalmanfilter. In order to improve the performance, it is necessary toperform post- processing. It has been accomplished by shadowand noise removal algorithms operating at the lattice whichidentifies object-level elements. The results of post-processing canbe used to detect object more efficiently. Experimental resultsand comparisons using real data demonstrate the superiority of the proposed approach which has achieved an average accuracyof 92% on completely novel test images.
 
 Keywords- Background subtraction; Background updation; Binary segmentation mask; Kalman filter; Noise remova; Shadow removal; Traffic video sequences.
I.
 
I
NTRODUCTION
 As computer vision begins to address the visualinterpretation of many problems in applications such assurveillance and monitoring are becoming more relevant. Ourmain goal is to identify the object from the multi modelbackground. For that we need to detect and extract theforeground object from the background image. Once objectsare detected, the further processing for tracking and activity islimited in the corresponding regions of the image. Severalfactors make on-road vehicle detection very challengingincluding variability in scale, location, orientation, and pose.Vehicles, for example, come into view with different speedsand may vary in shape, size, and color. Vehicle appearancedepends on its pose and is affected by nearby objects. In-classvariability, occlusion, and lighting conditions also change theoverall appearance of vehicles. Region along the road changescontinuously while the lighting conditions depend on the timeof the day and the weather.We present a system for detecting and tracking vehicles insurveillance video which uses a simple motion model todetermine salient regions in a sequence of video frames.Similar regions are associated between frames and grouped toform the background. The entire process is automatic and usescomputation time that scales according to the size of the inputVideo sequence. We consider image/video segmentation withinitial background subtraction, object tracking, and vehiclecounting, in the domain of traffic monitoring over anintersection.
 
The remainder of the paper is organized as follows: SectionII gives the overview of the related work. Section III describesthe architecture and modeling of proposed methodology forbackground elimination and object detection. Implementationand performance are analyzed in section IV. Section V containsthe concluding remarks and future work.II.
 
O
VERVIEW OF THE RELATED WORK
 Many works have been proposed in the literature as asolution to an efficient and reliable background subtraction.To detect moving objects in a dynamic scene, adaptivebackground subtraction techniques have been developed [1][2] [3]. Adaptive Gaussian mixtures are commonly chosen fortheir analytical representation and theoretical foundations. Forthese reasons, they have been employed in real-timesurveillance systems for background subtraction [4] [5] andobject tracking [6]. For foreground analysis [7] [8], a methodfor foreground analysis was proposed for moving object,shadow, and ghost by combining the motion information. Thecomputation cost is relatively expensive for real-time videosurveillance systems because of the computation of opticalflow. In [9], a work has presented on a novel backgroundsubtraction algorithm that is capable of detecting objects of interest while all pixels are in motion. Background subtractiontechnique is mostly used for motion pictures to segment theforeground object by most of the researchers [10] [11]. LiyuanLi, et al. [12] proposed foreground object detection throughforeground and background classifications under bayesian
275http://sites.google.com/site/ijcsis/ISSN 1947-5500
 
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 8, No. 4, July 2010
framework. In addition, moving object segmentation withbackground suppression is affected by the problem of shadows[6] [13]. Indeed, the moving object detection do not classifyshadows as belonging to foreground objects since theappearance and geometrical properties of the object can bedistorted which, in turn, affects many subsequent tasks such asobject classification and the assessment of moving objectposition. In this paper, we propose a novel simple method thatexploits all these features, combining them so as to efficientlyprovide detection of moving objects, ghosts, and shadows. Themain contribution of this proposal is the integration of knowledge of detected objects, shadows, and ghosts in thesegmentation process to enhance both object segmentation andbackground update. The resulting method proves to beaccurate and reactive and, at the same time, fast and flexible inthe applications.III.
 
P
ROPOSED WORK
 Block diagram of the algorithm used in proposed system isdescribed in Fig.1. The proposed Automated VideoSurveillance system seeks to automatically identify people,objects, or events of interest in variety of environment.Typically, these systems consist of stationary cameras placed inhighways. These cameras are integrated with, intelligentcomputer systems that perform preprocessing operation fromthe captured video images and notify human operators ortrigger control process. The objective of this real-time motiondetection and tracking algorithm is to provide low-levelfunctionality for building higher-level recognition capabilities.
Preprocessing &BackgroundSubtractionForegroundDetection
Shadow and NoiseRemoval
Object Detection
Frames f1,f2,.....fn
 
Figure 1. Block diagram of the algorithm.
 A.
 
Preprocessing
Preprocessing is the key step and the starting point forimage analysis, due to the wide diversity of resolution, imageformat, sampling models, and illumination techniques that areused during acquisition. In our method, preprocessing step wasdone by statistical method using adaptive median filter. Theresultant frames are then utilized as an input for the backgroundsubtraction module. In Fig.2, image
 I(x,y)
at time
is shown inFig.3 , the background image
 B(x,y)
at time
is shown.
Figure 2. Image at time
:
 I (x; y; t).
Figure 3. Image at time
:
 B (x; y; t).
To get the new estimated background, the following stepswere used for calculation.
Step 1:
 Estimate the background at time t using adaptivemedian filter method.
Step 2:
 Subtract the estimated background from the inputframe.
Step 3:
Apply a threshold, to the absolute difference to getthe binary moving objects hypothesis mask.Assuming that the background is more likely to appear in ascene, we can use the median of the previous n frames as thebackground model
)),,,(((),,(
i y x I  Median y x B
(1)
,|),,((),,(|
 
i y x I  Median y x I 
(2)where
}1,...,2,1,0{
ni
 
 B.
 
Foreground Detection
In this module estimated background and foreground mask images are used as an input for further processing .Thus, weuse grayscale image sequences as input. Elements of the sceneand the sizes of the traffic objects (vehicles and pedestrians)are unknown. The Foreground detection is done by usingaccumulative difference method, which is change-detectionbased on subtraction of a background image. It is necessary toupdate the background image frequently in order to guaranteereliable object detection. The basic idea in background
276http://sites.google.com/site/ijcsis/ISSN 1947-5500
 
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 8, No. 4, July 2010
adaptation is to integrate the new incoming information intothe current background image using a Kalman filter:
D M a M a B B
]*1(*[
21)1(
, (3)where
 B
represents the background model at time
 D
,
isthe difference between the present frame and the backgroundmodel, and
 M 
is the binary moving objects hypothesis mask.The gain are based on an estimate of the rate of change of the background. The larger it is, the faster newchanges in the scene are updated to the background frame. Inour approach, , they are kept smalland the update process based on Eq.(3) is only intended foradapting to slow changes in overall lighting.
Otherwise
 
 x Dif   x M 
,0|)(|,1 )(
. (4)Foreground detection is started by computing a pixel basedabsolute difference between each incoming frame and anadaptive background frame
 B
. The pixels are assumed tocontain motion if the absolute differences exceed a predefinedthreshold level. As a result, a binary image is formed whereactive pixels are labeled with a "1" and non-active ones with a"0". With the updated background image strategy usingKalman filter, we get the better foreground detection result.This is a simple, but efficient method to monitor the changesin active during a few consecutive frames. Those pixels whichtend to change their activity frequently are masked out fromthe binary image representing the foreground detection result
.
C.
 
Shadow Removal
Shadows appear as surface features, when in fact they arecaused by the interaction between light and objects. This maylead to problems in scene understanding, object segmentation,tracking, recognition, etc. Because of the undesirable effects of shadows on image analysis, much attention was paid to the areaof shadow detection and removal over the past decades andcovered many specific applications such as traffic surveillance.In this paper, 8-neighborhood gray clustering method is used todefine the precise shadow and remove it. The mean clusteringthreshold and the initial cluster seed of the gray are calculatedby the following equations.
2
)),(max()3 / 1(
ii
u y xG
, (5)where
),(
y xG
 
is a gray value of the pixel in
),(
y x I 
,
 
i
u
 
isthe mean of 
),(
y xG
. The initial seed
i
locates in the centre of 
),(
y xG
.The clustering starts from the seed
i
 
, and the point
i
P
is examined in turn. If at least one point in the 8-neighborhood of 
i
P
 
has been marked as a shadow region,standard deviation of 
i
P
 
is calculated by the followingequation.
2
)),((
iii
u y xPP
, (6)where
)),((
y x I P
i
 
is the gray value of I(x,y). If 
P
i
,
i
P
 
must be shadow point; otherwise, the point need not bemarked. The point
 
i
P
 
is checked constantly until no new pointis marked. At last all the marked shadow points are removed.
 D.
 
 Noise removal
Usually due to the camera noise and irregular objectmotion, there always exist some noise regions both in theobject and background region. Moreover, the object boundariesare also not very smooth. Hence a post processing technique isapplied on the foreground image. In order to remove the noisewiener low-pass filters a grayscale image that has beendegraded by constant power additive noise. It is based on anadaptive statistics estimated from a local neighborhood of eachpixel.
 E.
 
 Detection
After post-processing, the image is compared with the oneof the original frames (usually, the first frame). If the pixels areless than certain threshold, then they are ignored. Otherwise,they are replaced by the pixels of original image. This resultingimage will be consisting of the moving object ignoring thebackground and hence satisfying our requirement.IV.
 
I
MPLEMENTATION AND PERFORMANCE ANALYSIS
 This system was implemented on an Intel Pentium IV 280GHz PC. We have tested the system on image sequences ondifferent scenarios like traffic junction intersection, highwaysetc. Real life traffic video sequence are used to demonstrate thevehicle tracking from traffic video sequences using theproposed framework. All the videos chosen for vehicle trackinghave same light intensity and have been taken during day time.We convert the colour video frames to gray scale images.Automatic monitoring visual surveillance systemimplementation needs to detect vehicles using automaticbackground extraction. Background subtraction is the mainstep for vehicle detection. Fig. 4 shows number of successiveframes that are used to extract the background. Digital cameraused to take shots. The camera placed over the highwaydirectly. It shots eight frames per second.
1 2
a anda
1 2
0.1 0.01
a and a
277http://sites.google.com/site/ijcsis/ISSN 1947-5500

Activity (9)

You've already reviewed this. Edit your review.
1 hundred reads
1 thousand reads
Hammad Ansari liked this
poweredgeus liked this
See Siang Wong liked this
R. Rajesh liked this
suneetsaini liked this
Nikky Jun Lin liked this
Anhana Nguyen liked this

You're Reading a Free Preview

Download
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->