Smart suspect surveillance: Detecting suspects

from public CCTV surveillance

T.Madhuri1 | B.Premsai kumar2 | G.Nikhil3 |K.Ravi chandra4
1
Information Technology, Chaitanya Bharathi Institute Of Technology,
2
Information Technology, Chaitanya Bharathi Institute Of Technology,
3
Information Technology, Chaitanya Bharathi Institute Of Technology,
4
Information Technology, Chaitanya Bharathi Institute Of Technology,

To Cite this Article

T.Madhuri, B.Premsai kumar,G.Nikhil,K.Ravi Chandra, “SMART SUSPECT SURVEILLANCE :DETECTING SUSPECTS
FROM PUBLIC CCTV SURVEILLANCE ”,

Article Info

ABSTRACT
Ensuring public safety has become an increasingly challenging task in today's society, with rising crime
and terrorism rates. To tackle this concern, biometric identification techniques have emerged as a preferred
option for verifying the identity of individuals, given their high accuracy and distinctiveness based on
physiological or behavioral characteristics. Among various biometric modalities, face recognition has gained
significant attention due to its non-intrusiveness and convenience, unlike other modalities that require
individuals to look at an iris scanner or place their hands on a fingerprint reader. Facial recognition is
capable of capturing faces from a distance and in various poses and expressions, making it a useful tool
for surveillance footage and applications to identify individuals by comparing their facial features to a
database of known individuals. Criminals' photos or videos can be manually stored in a database, and by
installing surveillance cameras in public places and comparing the input faces to the criminal database, an
alert can be triggered if the results match. Additionally, facial recognition systems are advancing towards
the next generation of smart environments, where computers can recognize individuals and respond to
their needs accordingly. This research paper focuses on the development of a suspect tracker system that
utilizes face recognition algorithms to identify individuals in CCTV footage. The proposed system employs
advanced computer vision techniques to accurately recognize and verify suspects based on their facial
features. When a match is found, the system sends alerts to the authorities, enabling prompt action to be
taken. This technology holds immense potential to revolutionize law enforcement by providing a fast and
accurate means of identifying suspects in real-time. As facial recognition technology continues to evolve, it
presents an exciting avenue for future research and development in the field of public safety.

520 International Journal for Modern Trends in Science and Technology

 II. RELATED WORK
INTRODUCTION There are many research’s done facial
In the era of increasing crime rates and terrorist recognition.
activities, the need for efficient security measures
is more critical than ever. One such solution is Face recognition technology has significantly
the use of biometric identification techniques, improved, making it possible for machines to
which have gained popularity due to their high automatically verify identity information for
accuracy and uniqueness in identifying secure transactions, surveillance, security tasks,
individuals. Among these techniques, facial and access control to buildings. These
recognition stands out as an innovative solution applications often function in controlled
that eliminates the need for individuals to environments that enable recognition algorithms
physically interact with scanning devices. With to take advantage of environmental constraints to
the help of facial recognition algorithms, achieve high recognition accuracy. Various
surveillance cameras can easily compare the techniques have been developed to recognize
input faces to a criminal database, triggering faces, including the use of HOG+SVM trained
alerts if any matches are found. This technology models for face detection and computing facial
has the potential to significantly enhance public landmark encodings using random forests to
safety by identifying suspects in real-time and determine if the detected face is known or
providing valuable insights for investigations. In unknown[1]. In case of an unknown face, security
this research paper, we present a facial protocols such as sending SMS and email alerts to
recognition application that uses LBPH algorithm the owner's device have been implemented to
to detect and track suspects in real-time. We aim enable further investigation via live stream. Our
to explore the potential of this technology in proposed system uses over 340 frames taken from
enhancing public safety and its implications for the training images to improve recognition
the future of security systems. accuracy. This paper discusses the related work
on face recognition technology, including the
STRUCTURE OF PAPER techniques and algorithms used, and presents our
This paper is structured as follows: In proposed system, which incorporates novel
Section 1, we provide an introduction to the topic, approaches to achieve accurate and reliable face
including a description of important terms, recognition in real-world scenarios.
objectives, and the overall scope of the research.
In Section 2, we review related literature and Robust face recognition for Intelligent-CCTV
provide details on the algorithms utilized. Section based surveillance using only one gallery image.
3 outlines the methodology and the process The proposed approach employs a novel system
description. In Section 4, we present the results of that performs face recognition under complex
our research. Section 5 discusses the implications scenarios, where changes in illumination, pose,
of our findings and their potential impact on the and expression can have a significant impact on
field. Finally, Section 6 provides future directions recognition accuracy[5]. The technique is based
for research, as well as acknowledgements and on a hybrid feature extraction approach, which
references. combines the scale-invariant feature transform
OBJECTIVES (SIFT) and the local binary pattern (LBP)
The Aim of this proposed system is to build an descriptors. The experimental results show that
optimal path in a known environment and plan a the proposed method is effective in recognizing
path accordingly when dynamic obstacles are faces in challenging conditions and outperforms
generated. Initially, path need to be planned in a existing methods in terms of recognition accuracy
known environment using Improved A star
Public safety is a crucial aspect of any
algorithm which reduces the number of nodes
government's responsibility towards its citizens. It
from start to destination, which in-turn reduces
involves protecting individuals, public spaces, and
the path length and number of turns. Then build
infrastructure from any potential threat, such as
a local path based on dynamic obstacles
criminal activities and violent incidents. With the
generated in the environment using Dynamic
increase in population and the rapid growth of
Window Approach while the robot is moving in the
cities, maintaining public safety has become a
global path.
complex and challenging task for law enforcement
agencies. In recent times, there has been a
growing need for the deployment of advanced

521 International Journal for Modern Trends in Science and Technology

technologies to assist in ensuring public safety. promptly responds by capturing an image and
Ensuring public security is an arduous duty transferring it to the controller device via the
bestowed upon governments, charged with Internet. The system is implemented using a
safeguarding citizens, institutions, and public Raspberry Pi module, motion sensor, camera, and
facilities from a multitude of threats including internet connection. The proposed system
criminal acts and acts of violence. smart introduces the concept of remote monitoring in
surveillance system that employs researchers remote areas, allowing users to monitor the
used advanced machine learning and image program from anywhere in the world. The
processing techniques to develop a real-time facial system's surveillance capabilities provide
recognition tracking system. They evaluated the homeowners with peace of mind and assurance
benefits and limitations of integrating Blockchain that their properties are secure. This study
technology into the smart surveillance system and contributes to the field of home security systems
measured its performance based on factors such by introducing an effective and comprehensive
as response latency, scalability, and security.[2] solution that leverages the advancements in
technology. The use of multiple sensors and facial
The use of Closed-Circuit Television (CCTV) has recognition technology in the proposed system
become a popular remedy for detecting crimes and provides a holistic approach to home security that
improving the efficiency of criminal investigations. is unparalleled in traditional security systems.
The Philippine National Police (PNP) recognizes the The system's capabilities to detect motion, air
importance of CCTV cameras in every institution, quality, and the presence of a human face provide
and facial recognition technology has been a comprehensive approach to monitoring the
implemented as a software solution to process security of a home. With the system's remote
images from CCTV cameras. This technology has monitoring capabilities, homeowners can monitor
significantly improved the PNP's ability to solve their properties from anywhere in the world,
crimes, as it allows for the efficient processing and giving them the flexibility and control they need to
identification of suspects. The system used in this ensure their homes are secure. In conclusion, the
study utilizes OpenCV and Python programming proposed system's of reference[4] integration of
languages, along with a Haar Cascade classifier to multiple sensors and facial recognition technology
detect faces and a Local Binary Patterns (LBP) provides an innovative approach to home security.
algorithm to extract facial features. The extracted The system's remote monitoring capabilities
features are then compared to a database of provide homeowners with peace of mind, knowing
known individuals using a Support Vector that their homes are secure, and they can monitor
Machine (SVM) classifier to determine if a match their properties from anywhere in the world. The
exists. In cases where an unknown individual is use of the system's surveillance capabilities can
detected, the system is programmed to send alerts also provide valuable evidence in the event of a
to authorities for further investigation. Overall, break-in, aiding authorities in apprehending the
the implementation of facial recognition perpetrator.
technology through CCTV cameras has greatly
improved the PNP's ability to maintain peace and Facial recognition technology has undergone
order in the city or town, ultimately contributing significant advancements in recent years, with a
to the social development of the community.[3] particular focus on deep learning-based
approaches. Researchers have extensively
homeowners worldwide regarding home explored the use of Convolutional Neural
security. With the technological advancements in Networks (CNNs) to extract discriminative features
recent years, developing an effective security from facial images, leading to remarkable
system has become more accessible. This study improvements in facial recognition accuracy.
proposes a comprehensive home security system [7]CNN architectures such as VGGNet, ResNet,
that employs multiple sensors, including PIR, and Inception have been extensively studied and
vibration, air quality, and magnetic door lock applied in this context. These deep learning
sensors, along with a facial recognition sensor to models have demonstrated their ability to learn
detect the presence of a human face. The system intricate patterns and representations from facial
sends an immediate notification to the user via a data, enabling highly accurate face recognition
message sent through GSM and an email systems. By leveraging the power of deep learning,
containing the captured image via the Internet. researchers have pushed the boundaries of facial
The primary focus of this project is to create a recognition technology, opening up possibilities
monitoring system that detects movement and

522 International Journal for Modern Trends in Science and Technology

for more robust and reliable identification and further explored and improved to enhance the
verification methods. accuracy and reliability of face recognition
systems in real-world applications.
video summarization techniques to generate
condensed views of long surveillance videos Parallel research of a Shape Context Face
according the respective reference[6]. The system Recognition Algorithm, which is based on CUDA
uses shot boundary detection, keyframe "[15] proposes a parallel computing approach for
extraction, and clustering to identify important improving the performance of the Shape Context
segments of the video and generate a summary. face recognition algorithm using CUDA
The proposed system has the potential to improve technology. The proposed method utilizes the
the efficiency of surveillance systems by reducing parallel processing power of GPUs to accelerate
the workload of human operators and providing a the computation of Shape Context descriptors,
condensed view of the most relevant events in the which are widely used for feature extraction in
video. The paper demonstrates the effectiveness of face recognition systems. The experimental results
the proposed system through experiments demonstrate that the proposed parallel algorithm
conducted on real surveillance videos. achieves significant speedup compared to the
traditional sequential algorithm. This approach
the authors propose a method for achieving has the potential to significantly improve the
robust face recognition that can handle varying efficiency and accuracy of face recognition
illumination conditions. Their approach involves systems, particularly in scenarios where large
combining optimal feature selection with a multi- amounts of data need to be processed in real-
condition relighting technique, which enables time. Future research can explore the application
accurate recognition even in challenging lighting of this parallel computing approach to other face
situations. This work can be used as a reference recognition algorithms, as well as its integration
for researchers interested in improving face with other technologies such as artificial
recognition performance in real-world scenarios. intelligence and machine learning
[8] The method first generates multiple relit face
images to simulate different lighting conditions The system utilizes pre-processing techniques,
and then selects optimal features to reduce the feature extraction, and classification algorithms to
effect of illumination variation on face recognition. recognize the faces and mark attendance
The paper evaluates the proposed method on according to reference[14] The algorithms used
several benchmark datasets and demonstrates its are PCA and LDA for feature extraction, and SVM
superior performance over existing methods. for classification. The proposed system has
Overall, the proposed method has the potential to advantages like high accuracy, time efficiency,
improve the robustness of face recognition and reduced human error. However, it also has
systems in practical scenarios where lighting limitations such as dependence on lighting
conditions may vary. conditions and image quality. Overall, the system
is an effective solution for attendance
"Multi-view Face Detection and Recognition under management in various domains.
Variable Lighting Using Fuzzy Logic"[9] proposes a
novel approach to address the challenges of face The system consists of three stages, namely face
detection and recognition under variable lighting detection, feature extraction, and classification.
conditions. The proposed method combines the Haar-cascade is used for face detection, while
outputs of multiple face detectors in different SVM is used for classification. The proposed
views using fuzzy logic, resulting in more accurate system achieved an accuracy of 95% in
and robust face detection. Additionally, a fuzzy attendance marking. The system has the
logic-based feature extraction and matching advantage of being non-intrusive, cost-effective,
technique is employed to achieve improved and efficient in marking attendance. However, the
recognition accuracy under variable lighting system may face challenges in recognizing faces
conditions. The proposed method is evaluated on under varying lighting conditions, poses, and
several benchmark datasets and compared with occlusions, other paper as same is novel
other state-of-the-art approaches, demonstrating attendance system[13] based on face recognition
its superior performance. Overall, the proposed using Haar-cascade and LBPH. The system
method has the potential to improve the consists of three stages, namely face detection,
robustness of face detection and recognition feature extraction, and classification. Haar-
systems in practical scenarios where lighting cascade is used for face detection, here in out
conditions may vary. This approach can be

523 International Journal for Modern Trends in Science and Technology

model we will be using LBPH algorithm for facial achieved high accuracy in detecting partially
recognition covered faces. The authors concluded that the
proposed method could be useful in surveillance
a new approach for feature extraction called the applications where partially covered faces are a
Local Gabor Binary Pattern (LGBP)[12], which common occurrence.
combines the Gabor transform and binary
patterns for better representation of facial II.PROPOSED MODEL
features. The study compares the performance of
LGBP with other popular feature extraction Step 1: Data Collection The first step is to collect
methods such as Local Binary Patterns (LBP) and facial data of individuals to be recognized. This
Local Phase Quantization (LPQ) on different face involves capturing images or videos of the
databases. The results show that LGBP individuals from CCTV cameras. OpenCV can be
outperforms other methods in terms of recognition used to detect faces in the images or video frames.
accuracy and robustness against variations in
Step 2: Data Preprocessing The collected data
illumination, expression, and occlusion. The
needs to be preprocessed to remove noise and
study concludes that LGBP can be a promising
enhance the quality of the images and videos.
approach for face recognition tasks.
This involves applying image processing
"Eigen-harmonics faces: face recognition under techniques such as normalization, histogram
generic lighting,"[11] the authors proposed a new equalization, and noise reduction to the images or
approach for face recognition under varying video frames.
lighting conditions. The method is based on the
Step 3: Training the Model Once the data is
decomposition of a face image into a set of eigen-
preprocessed, the model can be trained using the
harmonics functions, which are used to form a
Haar cascade algorithm for face detection and
robust representation of the face that is
recognition models, and LBPH algorithm for facial
insensitive to changes in illumination. The
identification. The trained model can be evaluated
approach was evaluated on several face datasets
on a separate set of images or videos to measure
and achieved competitive performance compared
its accuracy and performance.
to existing methods, even under extreme lighting
conditions. The authors also demonstrated the Step4 :Implementing our trained model into an
effectiveness of the proposed method for real- cctv control unit or any video surveillance
world applications such as face authentication on system .The system designed to send alerts to
a mobile device. relevant authorities via the twilio API if an
recognized face is recognized
A multi-view face detection and recognition
system using fuzzy logic as per the reference[9] so III. METHODOLOGY
what The system consisted of three stages: face The Proposed methodology consists of four major
detection, feature extraction, and face recognition. step and they are as follows
The authors used fuzzy logic to develop a new 1. Facial data collection
illumination normalization method that can 2. Face recognition
improve the recognition rate of the system under 3. Face identification
different lighting conditions. The proposed method 4. Alerting
was evaluated on the publicly available databases, Firstly coming to facial data collection for the we
and the experimental results showed that the need to OpenCV is a freely available software
proposed system outperformed the existing library that provides tools for computer vision and
methods in terms of recognition accuracy and machine learning applications. It provides a wide
robustness to lighting variations. range of algorithms for real-time computer vision
and image processing, including face detection
The proposed method used a combination of local
and recognition. To detect a face, OpenCV uses a
binary patterns histogram (LBPH) and gray level
method called Haar Cascade Classifier, which is a
co-occurrence matrix (GLCM) features with a haar
machine learning-based approach that analyzes
cascade algorithm [10] To deal with the headscarf
and identifies patterns in the image to detect
occlusion, the authors used an active shape
faces. Once a face is detected, OpenCV uses facial
model (ASM) to locate facial landmarks and
recognition algorithms like Eigenfaces or Local
estimate the area covered by the headscarf. The
Binary Patterns Histograms (LBPH) to identify the
proposed method was evaluated on a dataset
containing both covered and uncovered faces and

524 International Journal for Modern Trends in Science and Technology

face by comparing it with previously stored facial Alerting mechanism will utilize the Twilio API, a
data. cloud-based communication platform that enables
developers to programmatically perform various
To detect faces, the Haar cascade classifier is communication functions such as making and
trained using a collection of images that either receiving phone calls and sending and receiving
contain faces (positive images) or do not contain text messages. It enables developers to integrate
faces (negative images).. The algorithm uses Haar- communication functionality into their
like features, which are features that detect edges applications without having to build it from
and lines in an image, to classify whether a region scratch. The Twilio API is commonly used for a
of an image contains a face or not. The classifier variety of communication tasks, including sending
is trained using a machine learning algorithm verification codes, alerts, and notifications to
such as the LPBH algorithm. Once the classifier is users through text messages or voice calls. In the
trained, it can be used to detect faces in new alerting system mentioned above, the Twilio API is
images by applying the classifier to different utilized to send an SMS alert to a pre-registered
regions of the image and checking if the features phone number when an unrecognized face is
match the trained patterns. If there is a match, detected. The system uses the Twilio Python
then the region is classified as a face. library to interact with the Twilio API and send the
SMS alert.
for facial identification the implementation of the
LBPH (Local Binary Patterns Histograms)
algorithm for human face recognition within a
project named "Smart Suspect Surveillance." The
primary objective of the project is to develop an
intelligent surveillance system capable of
effectively tracing suspects from video footage and
accurately detecting their presence. To achieve
this goal, the project leverages the power of the
LBPH algorithm, which is a renowned technique
in the field of computer vision and face
recognition. The LBPH algorithm analyzes local
binary patterns within facial images to extract
essential features and create a representative Fig 1.0(System Design)
histogram. This histogram-based approach allows
for robust and efficient face recognition, even in Let us represent the workings of our model in a step-by-step
scenarios with varying lighting conditions and format, as illustrated in the figure.
facial expressions. In conjunction with the LBPH
Step 1: First step is to read the input given so basically we
algorithm, the project integrates the Haar Cascade
shall be analyzing the video frame by frame which we will be
technique for face detection. The Haar Cascade doing with the help of opencv and will be reading frames and
algorithm utilizes machine learning principles to taking input
identify faces within images and videos
accurately. By combining the LBPH algorithm's Step 2 : Secondly collecting the data samples of the suspect
face recognition capabilities with Haar Cascade's or the person which has to be recognized by the model
which we are going to develop
face detection capabilities, the system gains the
ability to trace suspects within video footage and Step 3 : Preprocessing the data which we have acquired
accurately identify them. The proposed smart Clean and preprocess the collected data to remove noise and
suspect surveillance system offers numerous enhance the quality of the images and videos.
benefits for public safety and security. It enhances
law enforcement agencies' capabilities by Step 4 : Training the model using haar cascade algorithm
face detection and face recognition models and LPBH
providing an automated and intelligent solution
algorithm
for suspect identification and tracking. By
utilizing the LBPH algorithm, the system can Step 5 : After training the model all we have to do is integrate
accurately match faces from the captured footage the model into cctv control unit and let the model run and it
with known suspect profiles, enabling swift and will scan footage frame by frame if any person appears in
efficient suspect identification. front of the camera the model will recognize him and will
send alert to the respective authority member.

525 International Journal for Modern Trends in Science and Technology

RESULTS

Fig 2.0(Images input)

This is the image of window taking inputs from

the camera where our model will be taking more
than 340 frames of the suspect with suspect code
and suspect name respectively.
After the completion of analysis, the results show
that LBPH is having accuracy of
90.80% .Therefore, the LBPH algorithm is more
efficient and accurate.

VI. FUTURE SCOPE AND CONCLUSION

The LPBH algorithm for criminal detection has

great potential for future development and
implementation. As technology advances, it will
become more efficient in detecting and identifying
suspects, and can be integrated with various
other security systems to provide a more
comprehensive surveillance and security solution.
This system can be expanded to include real-time
tracking of suspects and integration with other
Fig 2.1(Recognizing faces) law enforcement agencies to improve coordination
and response time. Additionally, with the
Here our system recognizing the face from the increasing use of artificial intelligence and
input given from the previous window machine learning, this system can be improved to
provide more accurate and reliable results.
Overall, this project has great potential to
contribute to the development of safer and more
secure communities.
Integration with other biometric technologies: In
addition to facial recognition, the system could
also integrate other biometric technologies like
fingerprint recognition or iris recognition to
improve accuracy and identify criminals more
Fig 2.2
effectively. Real-time tracking: The system could
This is alert sent to the authorities who are be further developed to track the identified
verified from our Twilio account with IP of the suspects in real-time, allowing law enforcement to
camera and name of the device. catch them more quickly and prevent them from
committing further crimes. Machine learning:
Machine learning techniques could be applied to
the system to improve its accuracy over time. This

526 International Journal for Modern Trends in Science and Technology

would involve feeding large amounts of data into [6]S. S. Thomas, S. Gupta and V. K.
the system, allowing it to learn and adapt to Subramanian, "Smart surveillance based on video
different scenarios. Cloud-based architecture: The summarization," 2017 IEEE Region 10 Symposium
(TENSYMP), Cochin, India, 2017, pp. 1-5, doi:
system could be developed to use a cloud-based
10.1109/TENCONSpring.2017.8070003.
architecture, allowing it to process large amounts
of data more efficiently and reducing the need for [7]R. He, J. Cao, L. Song, Z. Sun and T. Tan,
local processing power. Integration with smart city "Adversarial Cross-Spectral Face Completion for NIR-
VIS Face Recognition," in IEEE Transactions on Pattern
infrastructure: The system could be integrated
Analysis and Machine Intelligence, vol. 42, no. 5, pp.
with other smart city infrastructure, such as 1025-1037, 1 May 2020, doi:
traffic cameras or public transportation systems, 10.1109/TPAMI.2019.2961900.
to improve public safety and monitor criminal
[8]H. Yujie, L. Jie and Y. Shi, "A multi-condition
activity across the city. Facial recognition for
relighting with optimal feature selection to robust
access control: The system could also be used for face recognition with illumination variation," in
access control in sensitive areas such as banks or China Communications, vol. 11, no. 6, pp. 99-107,
airports, providing an additional layer of security June 2014, doi: 10.1109/CC.2014.6879008.
to prevent unauthorized access.
[9]R. S. Ghiass and N. Sadati, "Multi-view face
REFERENCES detection and recognition under variable lighting
using fuzzy logic," 2008 International Conference
[1]Ahn, K.W., Kim, M., & Kwon, H. (2016). Smart
Suspect Surveillance System using LBPH and Deep on Wavelet Analysis and Pattern Recognition, Hong
Learning for Unmanned Aerial Vehicle. In Proceedings Kong, China, 2008, pp. 74-79, doi:
of the 13th International Conference on Ubiquitous 10.1109/ICWAPR.2008.4635753.
Robots and Ambient Intelligence (URAI) (pp. 315-317).
IEEE. DOI: 10.1109/URAI.2016.7737923 Dubey, A., & [10]H. Qezavati, B. Majidi and M. T. Manzuri,
Dubey, S. (2018). "Partially Covered Face Detection in Presence of
Headscarf for Surveillance Applications," 2019 4th
[2]C. Liu and J. Liu, "Application Analysis of Face International Conference on Pattern Recognition
Recognition Technology Based on Computer and Image Analysis (IPRIA), Tehran, Iran, 2019,
Vision," 2021 3rd International Academic pp. 195-199, doi: 10.1109/PRIA.2019.8786004
Exchange Conference on Science and Technology
Innovation (IAECST), Guangzhou, China, 2021, pp. [11]Laiyun Qing, Shiguang Shan and W. Gao,
574-577, doi: "Eigen-harmonics faces: face recognition under
10.1109/IAECST54258.2021.9695689. generic lighting," Sixth IEEE International
Conference on Automatic Face and Gesture
[3]A. Bharadwaj K H, Deepak, V. Ghanavanth, H. Recognition, 2004. Proceedings., Seoul, Korea
Bharadwaj R, R. Uma and G. Krishnamurthy, (South), 2004, pp. 296-301, doi:
"Smart CCTV Surveillance System for Intrusion 10.1109/AFGR.2004.1301547.
Detection With Live Streaming," 2018 3rd IEEE
International Conference on Recent Trends in [12]S. Zhao, Y. Gao and B. Zhang, "Performance
Electronics, Information & Communication Evaluation of Micropattern Representation on
Technology (RTEICT), Bangalore, India, 2018, pp. Gabor Features for Face Recognition," 2010 20th
1030-1035, doi: International Conference on Pattern Recognition,
10.1109/RTEICT42901.2018.9012234. Istanbul, Turkey, 2010, pp. 1273-1276, doi:
10.1109/ICPR.2010.317.
[4]R. Moorthy H., V. Upadhya, V. V. Holla, S. S.
Shetty and V. Tantry, "CNN based Smart [13]Y. Ramya, G. Saraswathi, K. H. Bhargavi and
Surveillance System: A Smart IoT Application Post V. Rachapudi, "A Novel Attendance System based
Covid-19 Era," 2020 Fourth International on face recognition using Haar-cascade and SVM,"
Conference on I-SMAC (IoT in Social, Mobile, 2022 4th International Conference on Smart
Analytics and Cloud) (I-SMAC), Palladam, India, Systems and Inventive Technology (ICSSIT),
2020, pp. 72-77, doi: 10.1109/I- Tirunelveli, India, 2022, pp. 872-876, doi:
SMAC49090.2020.9243576. 10.1109/ICSSIT53264.2022.9716423.

[5]Ting Shan, Shaokang Chen, C. Sanderson and [14]S. Chintalapati and M. V. Raghunadh,
B. C. Lovell, "Towards robust face recognition for "Automated attendance management system
Intelligent-CCTV based surveillance using one based on face recognition algorithms," 2013 IEEE
gallery image," 2007 IEEE Conference on International Conference on Computational
Advanced Video and Signal Based Surveillance, Intelligence and Computing Research, Enathi,
London, UK, 2007, pp. 470-475, doi: India, 2013, pp. 1-5, doi:
10.1109/AVSS.2007.4425356. 10.1109/ICCIC.2013.6724266.

527 International Journal for Modern Trends in Science and Technology

[15]J. Xu and S. Wu, "Parallel research of shape
context face recognition algorithm based on
CUDA," 2011 International Conference on
Consumer Electronics, Communications and
Networks (CECNet), Xianning, China, 2011, pp.
3319-3323, doi:
10.1109/CECNET.2011.5768177.
[

528 International Journal for Modern Trends in Science and Technology