19-07-2012, 11:31 AM
Embedded Smart Car Security System on Face Detection
Embedded Smart Car Security System on Face Detection.pdf (Size: 108.48 KB / Downloads: 311)
Abstract
In this proposed embedded car security system, FDS(Face Detection System) is used to detect the face of the
driver and compare it with the predefined face. For example, in the night when the car’s owner is sleeping and someone
theft the car then FDS obtains images by one tiny web camera which can be hidden easily in somewhere in the car.
FDS compares the obtained image with the predefined images if the image doesn’t match, then the information is sent
to the owner through MMS.
So now owner can obtain the image of the thief in his mobile as well as he can trace the location through GPS. The
location of the car as well as its speed can be displayed to the owner through SMS. So by using this system, owner can
identify the thief image as well as the location of the car
This system prototype is built on the base of one embedded platform in which one SoC named “SEP4020”(works at
100MHz) controls all the processes .Experimental results illuminate the validity of this car security system.
Key words - Face detection, Forward feature selection method, ARM (Advanced RISC Machine,),GSM,GPS.
I. INTRODUCTION
It consists of PC memory unit it stores the different
driver image. FDS (face detection subsystem) is used to
detect the face of the driver and compare it with the
predefined image. If the image doesn’t match then the
information is send to the owner through MMS. Owner
can trace the location through GPS. This system owner
can identify the theft image as well as the location of the
car.
Traditional car security systems rely on many
sensors and cost a lot. When one car is really lost, no
more feedback could be valid to help people to find it
back. We put forward the face detection technique to be
applied in car security system because this kind of
technique is effective and fast, and one alarm signal
could be given to make an alarm or “call” the police and
the host soundlessly with the help of other modules in
the system prototype.
Face detection techniques have been heavily studied
in recent years, and it is an important computer vision
problem with applications to surveillance, multimedia
processing, and consumer products. Many new face
detection techniques have been developed to achieve
higher detection rate and faster. We have introduced an
boosted cascade of simple classifiers using Haar-like
features capable of detecting faces in real-time with both
high detection rate and very low false positive rates,
which is considered to be one of the fastest systems.
In this embedded smart car security system, FDS
(face detection subsystem) aims at detect somebody’s
face in the car during the time in which nobody should
be in the car, for example, in the night when the car’s
owner is sleeping. FDS obtains images by one tiny
digital camera which can be hidden easily in somewhere
in one car. When FDS detects one face in alarm period,
one alarm signal will be sent to the control central of the
system. An alarm or a “silent” alarm will be triggered
according to the use’s settings. In silent alarm pattern,
no direct alarm will be made, but several modules are
working at inform owner and the police several
important data, for example, the precise location of the
car.
The GPS module obtains the precise locality by
parsing received GPS signal. The GSM module can send
the information out by SMS (Short Message Service)
Embedded Smart Car Security System on Face Detection
Special Issue of IJCCT, ISSN (ONLINE) : 2231–0371, ISSN (PRINT) : 0975–7449, Volume- 3, Issue-1
113
message, including real-time position of the “lost” car
and even the images of “the driver”.
Fig. I: Configuration of the low – cost extendable frame
work for Low cost embedded smart car security system
II. FACE DETECTION SUBSYSTEM
A. Cascade Detector
Face detection is to find whether there are faces in
one image or not and their positions, and it belongs to
“pattern recognition”, one hot study spot of computer
intelligence. Many methods have been put forward to
solve the problem. To speed up the system to meet the
real-time ability, we choose the cascade detector method
to be part of work bases, which has been proved to be
the nearly fastest method of all. A cascade face detector
uses a sequence of node classifiers to distinguish faces
from non-faces. The proposed face detection method is
based on a cascade of simple classifiers to handle each
part of one integral image”.
Fig. 2 : Cascade architecture with n nodes.
The main advantage of this kind architecture is its
detection speed: a cascade detector can detect faces
almost in real time. Every node is a classifier which
determines one image block contains faces or not by
several “features”. So, how to choose those features is
the key point.
B. The symmetric direct feature selection
We have proposed one new method called Forward
Feature Selection (FFS) to train node classifiers in a
cascade which is much faster than the original cascade
face detector using AdaBoost.
1) For node n, we are given the nth bootstrapped
training set, the minimum detection rate dn, and the
maximum false positive rate fn.
2) For every feature, j, train a weak classifier hj, whose
false positive rate is fn. Sort these weak classifiers
according to their detection rate and form a
classifier pool P with the first s weak classifiers that
have largest detection rates.
3) Initialize the ensemble H to an empty set, i.e. H
. t 0, d0=0.0, f0=1.0.
4) While dt < dn or ft > fn
(a) Find the feature k, such that by adding it to H, the
ensemble will have smallest asymmetric cost. The
asymmetric cost of the ensemble is defined as its
false positive rate plus times its false negative
rate, in which is the cost ratio.
(b) t t+1, H H U {hk}
© Calculate the new ensemble’s detection rate and
false positive rate.
5) The decision of the ensemble classifier is formed by
a majority voting of weak classifiers in H, i.e.
1 hj_Hh(j)
H(x) = 0 otherwise
Decrease if necessary.
A cost ratio function is used in this method, in
which a false negative costs more than a false positive,
and make the selection process as “asymmetric”. The
asymmetric feature selection method is 100 times faster
than the original symmetric feature in training process.
Because the car security system would be used
inside cars only, the background of images is all very
simple. So we can choose many pictures taken in cars as
non-face database to help training process produce
better classifiers which are more suitable for cars.