Social Distance Detection System

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 07 | July 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 1174
Social Distance Detection System
Tejas Potekar1, Manav Khimani2, Iffat Khatib3, Archana Chaugule4
1,2,3,4 Department of Information TechnologyShah & Anchor Kutchhi Engineering College, Chembur, Mumbai
------------------------------------------------------------------------***---------------------------------------------------------------------
Keywords— Social Distance, Object Detection, YOLO
V4, Python, OpenCV.
I. INTRODUCTION
II. LITERATURE REVIEW
Our study aims to estimate the number of COVID19
infections and the extent of their severity (e.g., shelter-in-
place and voluntary quarantine) to determine the amount
of hospital capacity required under social distancing. The
YOLO algorithm is used to detect objects.[1] YOLO is
designed to learn very general representations of objects.
Image processing helps to help the computer understand
the content of an image. It outperforms other methods like
DPM and R-CNN when applied to natural images [4]. In
OpenCV, programmers can find programming functions
primarily for image-processing applications. [5] Euclidean
distance allowsfor the calculation of the distance between
a set of points in Euclidean space given an incomplete set
of distances. Video tracking aims to associate objects in
successive frames of video. In terms of frame rate, it
works best when the objects move quickly. [7]
III. PROPOSED SYSTEM
Python, OpenCV, and Yolo V3 were used to build the
proposed system. Humans and objects are differentiated
by the system to increase accuracy. As soon as two or
more individuals are close together and violating social
distancing protocols, bounding boxes turn red. Also, the
system records how many violations were committed in
each area so disease hotspots can be determined.
Figure 1: Flowchart
Abstract: Contagious diseases are controlled by social
distancing. Social distancing is the concept that people
should physically distance themselves from each other,
reducing close contact, and with it, reducing the spread
of contagious diseases (like Coronavirus). We developed
a social distance detection system to assist officials in
identifying areas with the most violations of Covid-19.
By detecting how close two people get to each other, this
system keeps tabs on violations. By identifying high-risk
areas, officials can prepare and allocate resources to
areas that are more likely to have an increase in
infections.
It is the practice of minimizing contact between
potentially infected individuals and healthy individuals in
order to reduce the transmission of contagious diseases in
a population. This was accomplished through the
development of a "Social Distance Detection System"
(SDDS). The system separates human movement from
object movement for increased accuracy. By detecting and
recording all violations, the system can predict hotspots of
disease spread based on the total number of violations.
COVID-19, which causes the disease, spreads easily from
person to person at this time. Healthy people can become
infected when they come into contact with respiratory
droplets from sneezes and coughs of infected individuals.
Following social distancing can reduce the number of
infections. Pandemics can be better managed and
controlled using this system.

IV. YOLO V3
A real-time object detection system called YOLOv3 (You
Only Look Once, Version 3) recognises particular things in
films, live feeds, or still photos. To find an item, the YOLO
machine learning system leverages features that a deep
convolutional neural network has learned. The YOLO
machine learning algorithm has three versions, with the
thirdversion being a more accurate version of the first ML
method. Versions 1-3 of YOLO were developed by Joseph
Redmon andAli Farhadi.
Figure 2: Comparison of YOLO, Resnet 50 and ResNet 150
V. DBSCANALGORITHM
In essence, clustering analysis, also known as clustering, is
an unsupervised learning technique that separates the data
points into a number of distinct batches or groups, with the
goal of ensuring that the properties of the data points
within the samegroup are similar and that the properties
of the data points within different groups are, at least in
part, dissimilar. It consists of numerous various differential
evolution-based methodologies.
VI. DSFD
Compared to Haar Cascades, Dlib, MTCNN, and DNN, the
open-source Dual Shot Face Detector (DSFD) method
detects faces more accurately. It deals with three
important aspects of facial detection: feature learning,
progressive loss design, and anchor assign based data
augmentation. The WIDER FACE Face Detection
Benchmark also placed it first overall.
VII. OBJECTIVES
● This tool aims to determine where people are in
real-time and display a bounding box that turns red
when two people are dangerously close together.
● Secondly, we want to determine how many people
are not wearing masks. As a result, governments
can be alerted if a situation becomes extremely
dangerous by analyzing the movement of people.
● To prevent a widespread pandemic like Covid- 19,
it is also essential to curb the spread of contagious
diseases.
VIII. PROCESS FLOW
● Input: Live video can be captured via CCTV or
prerecorded video can be uploaded for analysis of
thevideo to run the system.
● Object Detection: Live cameras and prerecorded
video are used to detect objects using YOLO V4. To
ensure that the system only detects violations
between humans, we filter out the "PEOPLE" class
inthis step.
● In order to calculate the Euclidean distance, we use
Open CV to calculate pairwise distances between
the centroids.
● Checking whether the distance matrix for a person
has a value less than N pixels is done
IX. ADVANTAGES
● Detection of social distancing violations is
madeeasier by this system.
● Governments and concerned authorities can
alsouse it to record violations in different areas
● Detection of users not wearing a mask is done
which helps reduce the spread of contagious
diseases.
X. IMPLEMENTATION
A. Person Detection
An object detection system that is state-of-the- art,
YOLO (You Only Look Once) performs real-time detection
of objects. The bounding boxes of individuals in a video
frame are obtained using Version 3 (pre-trained on the
COCO dataset). Yolo v3 model was used to generate the
output layer applicable for person detection. A confidence
score of more than 0.5 is required to detect each
individual person. It is also possible to optimize the
processing speed using the YOLO v3-tiny model or
320x320 resolution. However, the accuracy of detection
will be reduced as a result.

Figure 3: Person Detection Implementation
After detecting the person, we track the centroid to
uniquely identify the individual and use a DBSCAN data
clustering algorithm imported from scikit-learn to
confirm the person's distance from the nearby
individuals. Using the above algorithm, we set the criteria
to determine whether or not social distance can be
maintained by passing the minimum safe distance in
pixels as a parameter. A bounding box is created with the
label = "unsafe" if the calculated distance between its
centroids is less than the criteria.
Figure 4: Social Distance Monitoring Implementation
C. Face Detection
Due to its superior performance over the state- of-the-
art face detectors, the Dual Shot Face Detector (DSFD)
networkis utilized throughout the project. Detecting faces
that are covered or with low resolution is not possible
with Common Face Detectors such as Haar- Cascades or
MTCNN. A large range of orientations can also be detected
using DSFD. The pipeline is a bit heavy, but the results are
accurate. Further, the mask classification model receives
the coordinates of the detected face.
Figure 5: Face Detection Implementation
D. Face Mask Classsifier
B. Social Distance Monitoring
In order to classify whether a detected face is
properly masked, we use a slightly modified ResNet50
model (with ground layers pre- trained on ImageNet).
AIn addition to the base layers, Sigmoid or SoftMax
classifiers are applied after some Average Pooling 2D and
Dense (with dropout) layers. A confidence score below
0.5 indicates the presence of a face mask, while a
confidence score above 0.5 indicates the absence of one.
A high-accurate and fast-training model emerged from
deploying the ResNet50 pre-trained network. About
96.5% accuracy was achieved by the Mask Classifier
model during the experiment.

Figure 6: Face Mask Classifier Implementation
E. Final Output
In the end the system was able to detect and identify
humans, measure the distance between these humans and
detect if the were violating social distancing norms. It was
also able to detect whether people were wearing a mask or
not. This system will hence prove very effective in
reducing the spread of contagious diseases.
Figure 7: Final Implementation
XI. CONCLUSION
This system will definitely detect violations of Social
Distancing on a large scale, allowing experts to take
precautions. By employing this system, both current and
future pandemics might be kept to a minimum. YOLO (You
Only Look Once) is a real-time object detection system that
provides state-of-the-art detection of objects. To obtain the
bounding boxes of individual persons in a video frame,
Version 3 (pre-trained on the COCO dataset) is used.
Person detection has been performed using the Yolo v3
model's output layer. A confidence score > 0.5 is used to
detect eachindividual. Speed optimization can be achieved
by using 320x320 resolution or using the YOLO v3-tiny
model. Despite this, the accuracy of detection will be
decreased.
ACKNOWLEDGEMENT
The project would not have been possible without Ms.
Archana Chaugule's support, patience, and belief in us, as
well as her flexibility regarding our work-release
schedules. The project would not be possible without
everyone's assistance. Additionally, we appreciate the
availability of computers in the lab and the help of our lab
staff members. This project would not have been possible
without the excellent facilities provided at our college.
REFERENCES
[1] "Keskinocak P, Oruc BE, Baxter A, Asplund J, Serban
N (2020) The impact of social distancing on
COVID19 spread: State of Georgia case study. PLoS
ONE 15(10): e0239798.
[2] “WHO. WHO Timeline—COVID-19 2020 [27 April
2020].
[3] “Social Distance Monitoring and Violation Alert
System in Public Places” A. Santhosh1 ,
[4] ”J. Redmon, S. Divvala, R. Girshick and A. Farhadi,
"You Only Look Once: Unified, Real Time Object
Detection” 2016 IEEE Conference on Computer
Vision and Pattern Recognition (CVPR), 2016, pp.
779- 788,doi: 10.1109/CVPR.2016.91.
[5] “OpenCV for Computer Vision Applications”
Mahamkali, Naveenkumar & Ayyasamy, Vadivel
[6] ”Euclidean Distance Geometry and Applications.
SIAM Review” Liberti, Leo & Lavor, Carlile &
Maculan, Nelson & Mucherino, Antonio. (2012).
[7] ”Visual Object Detection and Tracking using YOLO
and SORT” Akansha Bathija and Prof. Grishma
Sharma

Social Distance Detection System

More Related Content

Similar to Social Distance Detection System (20)

More from IRJET Journal (20)

Recently uploaded (20)

Social Distance Detection System