Moving Object Detection and Segmentation using Background Subtraction by Kalman Filter
Indian Journal of Science and Technology
Vol 10 (19) | May 2017 | www.indjst.org
2
Survey on this paper is persistent on techniques for
tracking objects in general and not to trackers custom-
ized for explicit objects. Kalman lter has been widely
used for tracking. e random variation and inaccura-
cies are observed over a time then Kalman lter is used
to approximation the unknown variable which is closer to
the measurement’s true values. is lter is used to esti-
mate the ambiguity of an expect value and it computer
a weighted average of the guessed value and considered
value. In varied domain of applications like control, navi-
gation, computer vision, and time series econometrics it
can be applied. e ler is based on the following three
parameters: object’s future locality prediction, decrease of
noise introduced by mistaken detections and facilitates
the process of organization of multiple objects to their
tracks. By taking the dierence between two frames we
can track the object then using domain knowledge along
with the displacement of scale and displacement vector
the motion of the object can have characterized. In this
paper we use Kalman lter for tracking object by assum-
ing variables are normally distributed.
Object tracking is the signicant problem in human
motion investigation. It is upper level computer vision
problem. Tracking engage corresponding detected fore-
ground objects between successive frames using dissimilar
feature of object like motion, velocity, color and texture.
In the surveillance system, locate the object in each frame
then nd its location in each frame trajectory of object is
generated. Object tracker may also supply the whole sec-
tion of the object in the frame at every instant of time
at the same it also provides correspondences between the
object instances between two frames.
We review some papers on background estimation,
background subtraction and object tracking. In
1
studied
at background estimation of a scene using a Kalman lter
approach. In the previous research introduces the concept
to track background intensity and adjust without human
intervention to modify in the scene. ere is two phase in
the algorithm in 1
st
step track update mean intensity and
in the 2
nd
step track standard deviation update. Updating
can take place according Kalman lter equation. Using
standard deviation, it modeled how the object transits
background and foreground. In result, the foreground
objects are basically detected from frame to frame with
particularly no sequential tracking. But a straightforward
improvement to the object detection would be to include
a simple tracker for the detected objects.
In
2,3
compare various background subtraction algo-
rithms for distinguish moving vehicles and pedestrians
in urban trac video series. In their experiments, they
vary the attributes in every algorithm to take dissimilar
precision working points. en evaluate them base on
how they diverge in preprocessing, background repre-
sentation, foreground nding, and data justication. Five
denite algorithms are tested on city trac video suc-
cession: frame dierencing, adaptive median ltering,
median ltering, mixture of Gaussians, and Kalman lter-
ing. Mixture of Gaussians generates the overriding results,
while adaptive median ltering suggests a straightforward
substitute with competitive performance.
In
4
proposed a spatio and temporal approaches for
Background representation by the region which can be
predicted by automatic process.
In
5
proposed an algorithm that is capable to contract
with both steady and unexpected total enlightenment
changes. ey had accessible an innovative algorithm
for background updating in support on Kalman ltering
method, which is strong to regular and jagged enlighten-
ment transform. e most important originality they have
establish, which make the algorithm strong to steady and
jagged illumination changes, is the evaluation of the total
illumination dissimilarity and its use as an exterior con-
trol of the Kalman lter. e testing had shown that the
algorithm spread out the diversity of development where
it cans eort successfully. Its most important drawback is
incapability to run dynamic background pixels.
In
6
proposed an algorithm which was able to dif-
ferentiate actions of interest in the land and imitation
events on the ocean face with enormously low false alarm
rate. Again, this technique was capable to contract with
the movement of the trees and outperformed obtainable
techniques. In their paper they proposed a technique for
modeling of dynamic scenes for the purpose of back-
ground foreground dierentiation and change detection.
e method relies on the deployment of visual ow as
a characteristic for change detection. In order to cor-
rectly utilize the doubts in the features, an original kernel
based multivariate concentration assessment procedure is