Authors :
Rumana Shaikh; Bharat Singh; Ramesh Singh Rajpurohit; Bharat Singh Rajpurohit
Volume/Issue :
Volume 6 - 2021, Issue 10 - October
Google Scholar :
http://bitly.ws/gu88
Scribd :
https://bit.ly/3GeC2K1
Abstract :
COVID-19 is currently threatening human life,
health, and productivity. Population vulnerability grows
as a result of a lack of effective remedial agents and a
scarcity of vaccines against the virus. As a result, social
distancing is considered an adequate precaution (norm)
against the pandemic virus's spread. The danger of viral
propagation can be decreased by avoiding physical contact
between people. This paper gives overview about how to
automate the task of monitoring social distancing using
input video and images, which is motivated by this idea. To
distinguish humans from the background, this
methodology employs the YOLO v4 model for detecting
objects. With respect to mean average precision (mAP),
frames per second (FPS), and loss values given by object
classification and localization in input frame, the YOLO v4
model's results are compared to those of other popular
models, such as FRCNN and SSD. The pairwise vectorized
L2 norm is computed using the 3D feature space obtained
by using the bounding box's centroid coordinates and
dimensions. Random forest regression is used to predict
violations using data collected from the Yolo v4 model.
According to the results of the experimental analysis,
YOLO v4 and random forest regression produced the best
results with balanced mAP and FPS score to monitor and
predict future possibility of infection using generated data.
Keywords :
m YOLO-v4, Object Detection, OpenCV, Random Forest regression
COVID-19 is currently threatening human life,
health, and productivity. Population vulnerability grows
as a result of a lack of effective remedial agents and a
scarcity of vaccines against the virus. As a result, social
distancing is considered an adequate precaution (norm)
against the pandemic virus's spread. The danger of viral
propagation can be decreased by avoiding physical contact
between people. This paper gives overview about how to
automate the task of monitoring social distancing using
input video and images, which is motivated by this idea. To
distinguish humans from the background, this
methodology employs the YOLO v4 model for detecting
objects. With respect to mean average precision (mAP),
frames per second (FPS), and loss values given by object
classification and localization in input frame, the YOLO v4
model's results are compared to those of other popular
models, such as FRCNN and SSD. The pairwise vectorized
L2 norm is computed using the 3D feature space obtained
by using the bounding box's centroid coordinates and
dimensions. Random forest regression is used to predict
violations using data collected from the Yolo v4 model.
According to the results of the experimental analysis,
YOLO v4 and random forest regression produced the best
results with balanced mAP and FPS score to monitor and
predict future possibility of infection using generated data.
Keywords :
m YOLO-v4, Object Detection, OpenCV, Random Forest regression
