SUSPECT TRACKING SYSTEM

Magesh kannan A,Manoharan S, Mukesh kanna G

Department of Computer Science and Engineering,

Adhiyamaan College of Engineering, Hosur,Tamil Nadu,India-635109

an hour to come up with an efficient and fast

detection and tracking algorithm.
1. ABSTRACT
Localization of human faces is
Face detection is the most popular area of considered as the primary and the initial
research in the vision of computer science. stage in the study of face detection. It
It has become an exploring domain in identifies the facial features by extracting
recent years due to the increase in the features from an image of the subject’s face
security demands and its law enforcement and analyses the relative position, size,
applications. It is the need of an hour to shape of the eyes, nose, cheekbones and jaw
come up with an efficient and fast detection of the detected person. Face localization can
and tracking algorithm. Haar cascade be referred to as the extraction of facial
algorithm is one of the efficient fast face features using a pattern recognition system.
detection and tracking algorithm. In this These obtained features are utilized to search
project haar cascade algorithm is used for for corresponding matching features in other
face detection and tracking. This project images. Open CV can be used for creating
will be implemented in public cameras such prototypes and systems.
located in various places. If the suspect pass
through any cameras, the suspect will get Facial recognition system spans across
caught by the camera. When a suspect get several applications such as psychology,
caught in any camera, the report will be biometric systems, immigration checking, e-
generated. With the help of this generated voting, banking domain and in the gaming
report, we can locate the suspect and the industry. A typical approach in face
police can easily track the suspect. This will recognition is video – content based
help us to track the suspect. This will searching. A retrieval system should return
reduce the workload for police with the all videos containing specific actors upon a
help of generated report. user’s request.
2. INTRODUCTION 3. EXISTING SYSTEM
Face detection is the most popular area Some facial recognition algorithms
of research in the vision of computer identify faces by extracting landmarks or
science. It has become an exploring domain features from an image. For example, an
in recent years due to the increase in the algorithm may analyze the relative position,
security demands and its law enforcement size, and/or shape of the eyes, nose,
applications. It is one of the most talked cheekbones and jaw. These features are then
domain in computer technology which is used to search for other images with
being used in a variety of applications that matching features. Other algorithms
identifies human faces in digital images. It is normalize a gallery of face images and then
used for automatic detection of a person with compress the face data, only saving the data
any image or video frame. It is the need of in the image that is useful for face detection.
A probe image is then compared with the
face data. One of the earliest successful effective as possible. The use of different
systems is based on template matching approaches such as neural networks and
techniques applied to a set of salient facial machine learning can lead to fast and
features, providing a sort of compressed face efficient solutions however, these solutions
representation. are expensive in terms of hardware resources
and power consumption. A possible solution
Drawbacks of the existing system: to this problem can be use of approximate
 Low efficiency. arithmetic. In many image processing
 More time to train suspect applications the results do not need to be
 Accuracy is not good. completely precise and use of the
 It consumes more time for processing approximate arithmetic can lead to reduction
the task. in terms of delay, space and power
consumption. Here the research is done to
4. LITERATURE SURVEY examine possible use of approximate
arithmetic in face recognition using Eigen
Various methods or experiments can be faces algorithm.
used for face recognition and detection
however two of the main include an 5. PROPOSED SYSTEM
experiment that evaluates the impact of facial
landmark localization in the face recognition The human face is recognized by the machine
performance and the second experiment using Haar-Cascade method, which is a
evaluates the impact of extracting the HOG technique, where faces are pre trained with
features from a regular grid and at multiple more sets of faces. Using this technique, a
scales. Here the study is to question of feature human face is identified and cropped. The
sets for robust visual object recognition. The cropped image will be converted into
Histogram of Oriented Gradients grayscale format. The conversion is done
significantly outperform other existing because the clarity of an image may vary
methods like edge and gradient based since the light condition can be low or high.
descriptors. We study the influence of each To avoid this variation in the lighting
stage of the computation on performance, condition, the image is converted into
concluding that fine-scale gradients, fine grayscale. Then these grayscale images are
orientation binning, relatively coarse spatial stored in a database for future recognition
binning, and high quality local contrast process. The training process is done using
normalization in overlapping descriptor ANN. After the training process is
blocks are all important for good results. completed, the machine will show the name
Comparative experiments show that though of the person along with a confidence level.
HOG is simple feature descriptor, the The name of the person who come in front of
proposed HOG feature achieves amazing the camera only if the face of that person is in
results with much lower computational time. the database otherwise it shows Unknown.
These processes are done by using OpenCV.
Face recognition is image processing
technique which aims to identify human faces Benefits include:
and found its use in various different fields
for example in security. Throughout the years  The response time is faster.
this field evolved and there are many  Accuracy is good.
approaches and many different algorithms  Training time is faster.
which aim to make the face recognition as
6. PROJECT DESCRIPTION 8. MODULES:
The system comprises of 3 major modules
The system is mainly focused to track the
with their sub-modules as follows:
suspect in a smart way to reduce the
workload for police.  Image Taking Module
 Learning Module
 The main objective of this application  Image Tracking Module
is to track the suspect
 This application detects and track the
suspect in a smart way. 8.1 IMAGE TAKING MODULE:
 Police will add the suspect Images. The face of a human is recognized by
 The Images will be trained. the machine by using Haar-Cascade method
 The trained images will help us to which is a technique where faces are pre
detect the suspect trained with more than 6000 sets of faces. By
 Easy to use and efficient using this technique, a human face is
computerized system.. identified and cropped in to 192 px X 112 px.
 Computerization can be helpful as Then that image will be converted into
means of saving time & money. grayscale format. This conversion is done
7.SYSTEM ARITECTURE because the clarity of an image may vary
according to the lighting condition since the
light condition can be low or high. To avoid
this lighting condition, the image is converted
into grayscale. It takes sixty images of an
individual person. The converted images is
then stored in the face database with their
respective names. It automatically stores the
hundred images and the name of the image
starts from 1 to 60. The numbers were
automatically incremented and will be stored
as image names. Those image files are stored
under the respective folder which was
The image is captured by using the initially created before the face tracking. The
camera and the face is detected by using Haar name of the person is passed to machine
- cascade method. Pre – processing step is through command line argument and it
performed where the face of the person in the creates a folder in the machine and the sixty
image is cropped separately in a specified images will be stored under the given name
pixel ratio. The cropped image is then of the person.
converted into grayscale format since the 8.2 LEARNING MODULE:
lighting condition may be vary and it may be
high or low. In order to avoid the variation in The data from the face database is
the lighting condition the image is converted taken and it creates a XML file. The face
into grayscale format. The converted image is database contains matrix of the gathered
then stored database to compare and match faces and it is created as an XML file. The
with the faces that may be helpful in the XML file contains the average faces of the
future. person’s label and under the label the matrix
of the images are stored.
 WEBSITES
8.3 IMAGE TRACKING MODULE: 1. https://en.wikipedia.org/wiki/Haar-
It follows the Face tracking process like_feature
from the starting to grayscale conversion.
Then the converted image is cross checked
with the faces in the database for any match. 2. https://www.tutorialspoint.com/pytho
If any match is found by the machine then it n_deep_learning
displays the name of that person otherwise it
displays person unknown.
9. CONCLUSION: 3. https://www.tutorialspoint.com/pytho
The easiest way to identify a person is n/index.htm
through their face. By using Haar-Cascade
method, the human faces are recognized
where the faces are pre trained with more
datasets of faces. The images are converted to
grayscale format. This conversion is done
because the clarity of an image may vary
according to the lighting condition since the
light condition can be low or high. Then these
converted images are stored in a database for
future recognition process. After all the
training process is done, the machine will
show the name of the person with a
confidence level using OpenCV.

REFERNCES:

[1] GARY BRADSKI AND

ADRIAN KAEHLER – Learning
OpenCV.
[2] MARK LUTZ – Programming
Python.
[3] NICHOLAS LOCASCIO AND
NIKHIL BUDUMA – Fundamentals
of Deep Learning.