SOCIAL DISTANCE DETECTOR
MINI PROJECT - 1
Abstract
In this COVID-19 period, Social distancing plays an important role in
preventing the spread of the virus. After the first phase of unlocking roads
are full of people in some of the metro cities, which results in the
violation of social distancing which will increase the spread of the virus.
In this paper we have studied 5 different Journal and Conference Papers
and discussed how we can use different technologies like Artificial
intelligence (AI), Machine learning (ML), python, Computer vision, etc,
to build a social distancing detection device which will monitor the
gathering and find the distance between different individual to check
whether the social distancing is followed or not. We can use this in busy
places like metro stations, offices and streets. After that, we will compare
the different methods and try to find the best and economical method.
Finally, we will talk about the problem that arises during this project and
how we will solve them.
Introduction
Almost 22 million people are infected with corona virus, out of which
2.77 million is in India. In India initially the recovery rate was 66-68%
which is increased and the new recovery rate is 73% which is a good
news, but social gathering and opening of schools and crowded places
will increase the spread of virus, Since march, more thank 35 million
people lost their job, global economy is in recession. GDP of china, Spain,
France and Italy fell by 36.6, 19.2, 21.3 and 17.5 per cent respectively.
Best way to prevent this virus is reducing public gathering and closing all
the college, malls and public transport but this will affect the economy
and increase unemployment, But we can open these services with
precautions and by applying social distancing, maintaining social
distancing with man power will only increase the risk of getting affected,
so, we will use wireless technologies to maintain the social distancing.
Fig showing the effect of social distancing
we will use neural network learning model and YOLO, R-CNN and DPM
for object detection and compare which will give the accurate results.
After detecting the image with the help of mathematical calculations we
will find the distance between two individuals and compare that distance
with the safe distance (6 feet), If the distance is more that the safe
distance that means they are following social distancing.
Fig shows the flow chart of social distance detector
To make sure that no individual will recorded multiple times, we will use
centroid tracking mechanism in which we will assign different ID’s to
each individual and monitor the distance between centroids. If we
observe that the distance between two individual is more than the safe
distance then we will mark that area in map and if we find more than 6
individual at same place violating social distancing then we will report
them.
Fig showing distance detection using centroid tracking mechanism
Comparison between different object detectors (YOLO, R-CNN and
DPM)
We get to know about a object detector by its mAP value, mAP stands for
mean average precision, which tells that how accurate does a detector
detect an object.
After training neural network with 3000 different images out of which
1000 images is used as weight for training, we came to know that the
mAP of YOLO, R-CNN and DPM is 57.8, 45.4 and 51.9 respectively,
and the time taken by YOLO is lesser than both R-CNN and DPM.
Fig MAp value of different detectors
Fig time taken by different detectors
YOLO speed and accuracy is better than others also we don’t have to
worry about the size of different object, It converts the object size to a
specific size, The size range varies from 320 X 320 to 608 X 608.
Social distancing detector using Block-chain
As we know that initially block chain is use as financial tool but now
days we can use it in health care system, The idea is to create an app
which is linked with user mobile number and will use GPS for tracking
the activity, every user will get some coins ,tokens and passes which they
can use for visiting public places, For example every user will get N
tokens each week and every token will carry T amount of time which
they can spent for their movement.
For places like school, malls, college and metro, we can restrict a certain
number of users and allow only limited number of users which will
reduce the rush and helps in maintaining proper social distancing.
Fig flow chart of social distancing using block chain
With the help of this we can even track the activity of infected patient, If
a user will get affected then we can easily track the places they visited
and other users they meet and interact.
The major drawback of this solution is, we can’t track someone who
don’t use mobile phone, although these days majority of people carry
smart phone but in village majority of family have one or two smart
phone which makes it harder to track every individual but this project will
work effectively if we try it on small scale like in office, stores and banks.
Normal V/S Bird eye view
If the input video is taken from some angle then for proper calculation
and working we have to convert that video into bird view, in bird view
we are seeing a video from the top, which make it way easier to calculate
the distance between centroids and convert that pixel distance into other
units (meter, feet, etc). For converting a normal video into bird view we
have to pick four points in perspective view and then convert that four
square points into bird view. Another thing we can do is by applying
unitary method we can calculate how many pixels are there in safe
distance (6 feet), then convert the end result.
Comparison between different methods
Ref.
No Methodology Method Result
Detection of people Used in large scale if
using YOLO object implemented properly
detection and and provide
1 Python and YOLO calculating the distance employment to those
object detector. between the centroids who will monitor the
using python. device for maintaining
social distancing
With the help of app Highly effective and
giving token and economical easy to use
movement pass every but user need to have
2 Block chain week for accessing smart phone.
small but important
place like bank and post
office.
Detecting object using Virtual perceptive view
YOLO and generating is highly overlapped and
virtual perceptive views minimizing over sized
3 Neural network and with the help of boxes are necessary.
bounding boxes omnidirectional
image.
Training model on
ImageNet and COCO
data set.
Transmitting radio Economical and can be
signals with wireless used in small areas,
access point (AP) to accuracy range up to 5-6
4 Wi-Fi communicate with other meters.
devices in the range. High privacy risk.
Wireless Unless like Wi-Fi can
communication with be used both indoor and
rang of 2.4 to 2.485 outdoor, low cost and
5 Bluetooth GHz. able track the movement
Automatically find and of infected people
connect to the nearby
devices, other devices in
range means violation
of social distancing.
Conclusion
Social distancing is one of the most effective way to prevent the spread of
virus and with the help of technology like python, block chain, AI, Wi-Fi,
etc, we can stop the violation of social distancing.
In this survey we have seen that how we can maintain social distancing
with the help of technologies. We discussed about different type of object
detectors and their advantages over others then we try to understand why
bird view is more convenient and how we can track the infected and
maintain social distancing with the help of commonly used technologies
like Wi-Fi and Bluetooth. We compared 5 different journals and try to
understand what is best for indoor and outdoor conditions considering the
fact that it is economical and easy to use.
References
1 OmniDetector: With Neural Networks to Bounding Boxes
Chemnitz University of Technology Chemnitz, Germany
Roman Seidel, André Apitzsch, Gangolf Hirtz
2 COVID-19: Prolonged Social Distancing Implementation Strategy
Using Blockchain-Based Movement Passes
Chandan Garg & Agam Bansal & Rana Prathap Padappayil
3 Pandemic Security System for Police using Neural Networks
Pimpri Chinchwad College of Engineering and Research , Pune
K.B Wane , Dr. Rahul G. Mapari and Ajin Abraham
4 Navaneeth Bodla, Bharat Singh, Rama Chellappa, and Larry S. Davis.
Soft-nms - improving object detection with one line of code. In IEEE
International Conference on Computer Vision, ICCV 2017
5 Enabling and Emerging Technologies for Social Distancing: A
Comprehensive Survey
University of Technology Sydney, Australia
University of Luxembourg, Luxembourg
