Sei sulla pagina 1di 7

LIVE FACE: RECOLLECTION SYSTEM IN LIVE

STREAM
Julie J​[i]​,Sreeram G​[ii]​,Eniyan S​[iii]​,Srinivasa Raghavan M​[iv]
[i]​
Assistant Professor ,Department Of Information Technology ,Sri Sairam Engineering College
julie.it@sairam.edu.in
[ii]​
Student ,Department Of Information Technology ,Sri Sairam Engineering College
sreeramgunasekaran@gmail.com
[iii]​
Student ,Department Of Information Technology ,Sri Sairam Engineering College
seniyan1sfw@gmail.com
[iv]​
Student ,Department Of Information Technology ,Sri Sairam Engineering College
msraghavmsr@gmail.com

I. Abstract~ has attended an event,they are marked as


In our Live face Recollection present in the back end database.
systems, the facial pattern of a person is
being noted during Livestream using
machine learning algorithms. Faces can be
detected through our camera which then
reveals their identity on the screen. Faces
can be identified through the collection of
the previous image of the person and
checking with spatial structure and
​ igure 1 Simple Face Identification System
F
arrangement. Hence we can then able to
The manual attendance record system
track the person of our interest.
is not efficient because it requires more time
Simultaneous detection and identification of
to arrange the records and also processing the
each and every individual within the frame
average attendance of each student. ​Hence
video boundary can be made possible.
there is a need for an alternative for existing
Hence the image is then used to detect faces
biometric system ​to avoid fake and dummy
visible in the camera and then training the
authentication​.Face Recollection is then being
images and are stored in a database with a
a subject of topic where the individual faces
name given by the user.
are being stored and identified.It not only
works well for attendance management
Keywords:Machine Learning,Localization,
systems but also for security and surveillance
Threshold,Feature Extraction,LBPH
systems and ​can be integrated with existing
Algorithm
biometric systems​.Here the suspects facial
identity are being stored in a database
II INTRODUCTION
maintained by the police department and if
Most of the existing biometric system
found anywhere in the surveillance camera
are used as Attendance Management system
installed by the police the suspect identity is
to hold the record of the number of times a
being revealed and the police agencies can
person attends a place or institution.If a person
easily find out the suspect who went on recognition/classification and clustering. The
missing.Several facial recognition algorithms framework is evaluated on human faces, by
are being present but an appropriate algorithm verifying whether two faces belong to the
must then be chosen which ensures better same person and grouping faces that belong to
recognition rate and ability to produce same person like in Google Picasa.It focuses
accurate results even in the worst lighting on triplet loss as the main contribution.
conditions
IV A : PROPOSED MODEL
III: LITERATURE SURVEY In the proposed system facial
[​ 3]​proposed an Automated Face recollection system during the live stream, we
Recognition System which is implemented can able to identify each and every individual
Time & Attendance application which will who is being captured during the live stream
detect the human faces and then it is resized to which will show a brief detail about each and
the required size and processed by using every person who is spotted in that
Viola-Jones Algorithm. Finally, the resized frame.Local Binary Patterns Histograms
face samples are recognized using a simple algorithm is being used for Facial recollection
Principal Component Analysis based process. From the input face obtained from the
dimensionality reduction techniques. Once camera, we can then rank the similarity of the
recognition is done, the attendance will be faces with existing one present in the
automatically updated in the Microsoft Excel database.The probability of the closest
Sheet along with the candidate name, their matching of the person can also be predicted
date and time of entry.Then HTML file is with their corresponding percentage of facial
created and the attendance records will be recognition. This mechanism will identify
automatically updated in the Application each and every individual who are all within
system to make sure only an authenticated the boundary of the frame with their respective
user can access the attendance sheet.They are details. The identification is done with the
then integrated into an Auto Attendance help of sensing of the individuals face nuances
Management System. like shades and shapes of the organs present in
[6] ​proposed a solution where the the face.The proposed system is based on the
camera will take an image of the entire class ISO Standard ​ISO/IEC JTC1/SC37 which is
room and detects each and every individual used to store and access the dataset to
face of the candidate within the video frame recognize the individual.
and updates their attendance in the attendance
database. The image is going to be captured USE CASE 1:
twice-once at the start and once at the end to The above use case describes the basic
confirm that the student has attended the entire working of Recollection system.
class.
[2] ​states about embedding learning

framework for face verification,


A Camera is being displayed at a
distance from the entrance to capture the
incoming facial images of the person. The
captured image may be of different size based
on the resolution and position of the camera. A
fixed resolution must be set for each captured
image dataset. Auto-resizing may sometimes
result in poor performance of the recollection
system​.

USE CASE 2:

Figure 2: Sample Live Face System Working

V A: METHODOLOGY
Figure 3: Internal Working Of Live Face System
The working of ​Figure 2 ​are
performed in several cases and at totally
V.B PREPROCESSING:
different locations. Therefore, it needs some
The detected face is extracted and is
quite transmission of data. Also, data
then subjected to preprocessing step. This
compression could also be needed to reduce
preprocessing step involves with histogram
transmission bandwidth.The working of LBPH
levelling of the extracted face image and they
algorithm is explained in [7].Face
are resized. The histogram-levelling is the
Recollection module is classified into two
most common histogram normalisation
phases enrollment phase and recollection
technique. This improves the distinction of the
phase.
image because it stretches the range of
i)Enrollment Phase: To add new datasets in
intensities in a picture by creating it more
the Face database and then can be recollected
clear. ​Figure 4 represents the standard image
later
preprocessing technique with appropriate
ii)Recollection Phase: Recollects all the
steps.
datasets which are being stored and then
performs LBPH Algorithm on it.
V.B.1 IMAGE RESIZING:
When the incoming face is detected,
V C: IMAGE CAPTURE
they are resized to 92 * 112 resolution. This
resolution depends upon the environment and bounding box is then being drawn around it to
lighting condition. achieve liveness detection.
Various algorithms are projected for
V.B.2 CONTRAST INTENSIFICATION: face detection like Facial geometry-based
For better recognition rate, we have to strategies,different types of Feature-Invariant
use the Min-Max Linear Contrast Stretch strategies, Machine learning strategies.In Face
Enhancement Technique.Here in the Contrast detection using Haar Cascades to recognize
Intensification technique the resized image is the faces.
being transformed to a grayscale image where
the facial parts are being identified from the V.E DATABASE PROCESSING:
image Here the image is being captured,
processed and stored in Face Database
It is a Database collected of frontal face
images taken under different days, with
webcam or camera. The processing module in
face recognition may convert the initial
information (or perhaps the degraded data
when compression and expansion) into a
​ ​Figure 4 : Image Pre Processing
feature vector, to preserve all the discriminant
data that can be accustomed to distinguish two
V.C IMAGE SEGMENTATION
different people, and removing all redundant
In the Recollection system
data.
segmentation of the image are based upon the
Threshold. The Threshold values are used in
V.F FEATURE EXTRACTION:
both processing the datasets and to recognize
The features from the images could be
the images based on the lighting support of the
extracted by the EigenFace Face Recognizer
environment.Here the Simple Threshold for
present in the OpenCV.The input image
processing of the input datasets.The Adaptive
removes unnecessary extra information which
threshold are used in the matching of the
is not required for classification.It simplifies
human face and comparison of the human
the image by extracting the necessary features
face.Cv.adaptivethreshold is built in function
contained in the image and avoids the rest of
in OpenCV for image segmentation.Threshold
the features. ​Equation (1) is applied to
value can be assigned manually based on
different environmental lighting conditions. describe the variation of the current pixels to
its neighborhood pixels.
n
V.D FACIAL DETECTION:
A proper and efficient facial detection
rule always enhances the performance of face
D=
√ ∑ (hist1i –hist2i ) 2
i=1

where D is LBP operator


(1)

recognition systems.Localization is done ​if the


face of the person is detected on the screen a
bounding boxes.

VI: PRECAUTION:
● A Single Individual must be present in
a dataset for efficient face recognition
● Distance between camera and person
should be 70 cm to 110 cm.
● Environment lighting condition should
be between 250 lux to 740 lux.
● Training data of a person should be
Figure 5: Histogram Extraction error free that is only his/her image
should be there in a dataset.
V.F.1 FEATURE VECTOR
From ​Figure 5 facial part is divided ● Training data of a person should be
into multiple smaller regions.The Histograms free from image filters.
are extracted from those regions and they are ● A WebCam is suggested for better
combined to form a unique feature vector.A resolution
Feature Vector is a design matrix in which
each row in a matrix is termed as a feature VII:REPORTS & DISCUSSIONS:
vector.It gives an efficient representation of a) Input Datasets
the face and used to find similarities between The input datasets are then being
images. stored in the computers with around 200
datasets of each and every person are being
stored in Face Database.Dataset with different
facial expression. From figure.Each individual
pictures are stored individually in a separate
folder to differentiate each person accurately
in a Face Database
Figure 6: Feature Vector

V.G FACE RECOGNITION:


In the Face Recognition phase the
LBPH algorithm tries to find out the local Figure 7: Test Images
structure of each image by comparing each
pixel with their surrounding pixels b) Face Before Training
step-by-step according to ​Equation In the training phase of the system,LBPH
(1).Localization is then achieved if the face Recognizer is then being applied to store the
histograms of the input datasets.The Live
classifier is found along with their.The name
Input Face is then being compared with the
of the facial datasets are then recollected if the series of already present facial datasets.If the
appropriate dataset matches then the name of live input face does not match with any of
the individual will be displayed with the them ​UNKNOWN ​is specified for the
incoming individual the integrity and truthfulness of a data which
also automate the process of identification.
Live detection of individuals will be used in
many fields of application in daily activities.
This will enhance the automation process
which is useful for saving a valuable amount
of time for especially counter-terrorism
activities instead of verifying the information
received each and every time. The future holds
the promise of automating surveillance system
which is enabled by Face recollection.

IX: FUTURE ENHANCEMENTS:


Figure 8: Output Before Training the dataset of Future enhancements can be made in
the person this project by revealing some additional
c) Face After Training information about a person such as their
Facial Recollection is being done by address or domain of the organisation where
the person is placed.The distance between the
using OpenCV face classifiers and webcam and the target individual can be
Facial Trained Dataset.Name of the reduced to avoid mismatch of faces.
individual is being displayed on the
screen with bounding boxes localizing X: REFERENCES
the face of the individual. This system 1. Shuo Yang, Ping Luo, Chen Change Loy, Member,
provides an accuracy of around 70% in IEEE and Xiaoou Tang, Fellow,IEEE
“Faceless-Net: Face Detection through Deep Facial
the lowlight environment. Part Responses”, IEEE Transactions on Pattern
Analysis and Machine Intelligence, 2017.
2. Florian Schroff, Dmitry Kalenichenko, James
Philbin, Google Inc “FaceNet: A unified embedding
for face recognition and clustering”, Computer
Vision and Pattern Recognition (CVPR), 2015.
3. Aparana Behara and M.V. Raghu, “Real-Time Face
Recognition System For Time And Attendance
Applications”, International Journal of Electrical,
Electronics and Data Communication,June 2013
4. Automated Attendance Management System Based
On Face Recognition Algorithms Shireesha
Chintalapati, M.V. Raghunadh
5. Facial expression recognition based on a mlp neural
network using constructive training algorithm Mayank
Agarwal, Nikunj Jain, Mr. Manish Kumar and Himanshu
Agrawal
Figure 9: Recognized image 6. “Attendance System based on Face Recognition using
Eigen face and PCA Algorithms” Priyanka Wagh,Jagruti

VIII: CONCLUSION Chaudhari,Shweta Patil,Roshani Thakare


7. "Local binary patterns and its application to facial image
Though Verification system for analysis: A survey” Di Huang,Caifeng Shan,Mohsen
physical and electronic access security is Ardabilian,Yunhong Wang,Liming Chen
available today, Recollection system ensures
8. “A Survey of Face Recognition Algorithms and Testing
Results”William A. Barrett
9. “Technological and Social Issues in the development and
deployment of Facial Recognition Biometric Systems”
Shaukat Ali Khan Master Thesis Computer Science.
1. “ Comparison of face Recognition Algorithms on Dummy
Faces”Aruni Singh, Sanjay Kumar Singh, Shrikant Tiwari
The International Journal of Multimedia & Its
Applications (IJMA) Vol.4, No.4, August 2012

Potrebbero piacerti anche