COVID-19 is an infectious disease that has spread globally, and the best
way to slow down transmission is to maintain a safe distance. Due to the COVID-19
spread, social distancing has become very vital. Furthermore, the formation of groups
and crowds cannot be left unseen. Even when the necessary regulations have been
implemented by governments worldwide, people tend not to follow the rules. We
wanted to make it possible for authorities in areas like schools, universities, industries,
hospitals, restaurants, etc., to monitor people breaking social distancing rules and take
appropriate measures to control the virus from spreading. To monitor and control the
crowd, society requires a system that does not put other people's lives at risk.
Therefore, it is critical that we stop it from spreading further. Initially, the government
imposed a lockdown to control the spread of the virus. Due to the lockdowns, the
economy had experienced some negative effects. Due to the economic slowdown,
people were allowed to go out and carry on with their regular tasks, leading to
crowding in many places, intentionally or unintentionally. The research work aims to
make a crowd detection and alert system in public places like hospitals, schools,
universities, and other public gathering events. The proposed idea has two modules; a
deep CNN CrowdNet people counting algorithm to detect the distance between humans
in highly dense crowds and an IoT platform for sending information to the authorities
whenever there is a violation. Image processing is carried out in two parts: extraction of
frames from real-time videos using YOLO CV, and the second is processing the frame
to detect the number of people in the crowd.
The crowd counting algorithm, along with the vaccination, will enforce safety rules in
people-gathering places and minimize health risks and spread. The image processing
YOLO model mainly targets people not following social distancing norms and standing
very close by. The data for the violations are sent online to the IoT platform, where the
value is compared to a threshold. The platform aids in sending alerts to the concerned
authorities in case of significant violations. Warnings are sent through e-mail or
personal messages to the concerned authorities and the location. This model prevents the presence of an official to check whom all are violating the
rules. There is no need for human intervention and risking their lives; direct messages
can be sent through the IoT platform to authorities if there is a crowd formation. Data
analytics can help find out the peak hours of crowding and help control the crowd
much more efficiently. CrowdNet, a deep CNN algorithm, will estimate the number of
humans in a given frame to classify the locations where most people communicate and
check whether the safe distance is not reached and the number of times it is not
reached. Our system sends the number of people available in the frame at that moment
and whether they are maintaining social distancing or not. The Deep CNN algorithm
will filter the objects by capturing high-level semantics required to count only the
humans and calculate the distance between the humans alone. The base neural network
is Alexnet to estimate whether it is safe or not and then send it to the respective
authority. This proposed idea using CrowdNet CNN and IoT combination will help
find out peak hours of crowding and help control the spread of the disease during social
distance violations without human intervention. Thus, social distancing in public places
is automated using the real-time deep learning-based framework via object detection,
tracking, and controlled disease.
Keywords: COCO, Convolutional Neural Networks (CNN), CrowdNet, Social Distancing, Ubidots, YOLO.
