REAL-TIME SOCIAL DISTANCE SURVEILLANCE SYSTEM
Laman Hasanli, Parvana Guliyeva, Sona Mehdizada
Baku Higher Oil School
Baku, Azerbaijan
sona.mehtizada.std@bhos.edu.az
Supervisor: Ph.D Associate Professor Kamala Pashayeva
Keywords:
YOLOv3, Convolutional Neural Network, video processing, object detection
With its devastating spread, the ongoing COVID-19 coronavirus
outbreak has caused a global calamity. Because there are no vaccines
THE 3
rd
INTERNATIONAL SCIENTIFIC CONFERENCES OF STUDENTS AND YOUNG RESEARCHERS
dedicated to the 99
th
anniversary of the National Leader of Azerbaijan Heydar Aliyev
164
which would cure the virus available in the current circumstances,
social separation is considered an adequate precaution against the
spread of the pandemic virus. This paper describes a method for
detecting social distancing using deep learning to assess the distance
between people in order to reduce the impact of the coronavirus
epidemic. By analyzing a video feed, the detecting tool was created to
warn people to keep a safe distance from one another. A pre-recorded
video of pedestrians walking was used to validate the proposed
strategy. With the accuracy of 95%, the outcome demonstrates that
the suggested method is capable of determining the social distancing
measures between many participants in a video.
Social distancing refers to strategies for preventing the spread of
a virus by limiting human physical contact in public locations (e.g.,
shopping malls, parks, schools, universities, airports, and workplaces),
avoiding crowds, and maintaining a safe distance between individuals.
If applied early on, social distance can play a critical role in halting
virus propagation and preventing the pandemic from reaching its peak.
Social distance has been shown to minimize the number of infected
patients and the strain on healthcare institutions. Therefore, in today’s
world an accurate social distance detection system is a necessity
rather than being preferred.
The pre-trained model based on the YOLOv3 method was utilized
to identify pedestrians using the video frames from the real-time
camera as an input. Later, the video frame was transformed to a top-
down view for determining distance values in the 2D plane. Any pair
of persons having distance less than predefined threshold value in the
display will be depicted with a red box on views. An approximation of
physical distance to pixel is utilized to approximately estimate social
distance violations between people. To determine whether the
distance value exceeds the minimal social distance, a violation
threshold is determined.
The model is created using the image and video captured by the
camera. The camera is set up to capture at a fixed angle, and the video
frame's view has been converted to a 2D bird's eye view to precisely
estimate the distance between each object detected. The people in the
picture are assumed to be leveled on the horizontal plane for simplicity.
Then four locations on the horizontal plane are determined, and the
scene is converted into a bird's eye view. Then, using the bird's eye
view, each person's position can be estimated. The CNN model was
|