COVID-19 ANALYSIS AND FORECASTING
INTRODUCTION
The covid-19 virus affected many countries across the globe in different ways. Almost all countries announced lockdown for more than 2 months and because of this, there is a strong impact on people’s wealth and also on their mental health. The government of India decided to open the lockdown and announced some strict rules that every person should follow like social distancing, usage of masks, etc.., Generally in public places like malls where a lot of people gather at once and there is a need to ensure that those people follow the rules set by the government. Monitoring every person manually is a time-consuming task and this process should be automated to ensure that every person is safe. And when it comes to any medical emergency like if a person suddenly falls in a public place then this incident should be known by the nearby hospitals in very little time. And this process should also be automated to ensure that the person is safe and decrease the reaction time from the hospital’s side.
And also to plan for the future the government of India needs information regarding future covid-19 cases, deaths, recoveries, and active cases state-wise.
We can make use of the CCTV footage in malls and also the information provided by the government about the covid-19 cases, deaths, and recoveries to solve the above problems.
So I and my teammate Vvs Manideep took this as a problem statement for our capstone project and we developed a model for each problem.
Our approach:
Social — distancing :
We developed a model that can detect whether people are maintaining social distancing or not. The model will count the no. of persons in an image using HOG (Histogram of oriented gradients) and SVM, and for detecting whether those people are maintaining social distance we used a centroid tracking algorithm to calculate the distance between those persons. We are using 3rd party software — UBIDOTS for displaying the results.
Usage of masks :
We developed a model that can identify whether the people in an image are wearing masks or not. We used the Haar cascade algorithm for detecting the face of a person and after that, for detecting the mask we used the transfer learning method because for masks there very fewer images available. So we used the Mobile net v2 pre-trained model for detecting the mask.
Fall detection :
We developed a model that can detect a situation like a person falling down suddenly due to medical illness. For fall detection we used the chainer pose estimation algorithm and after that, we are using google API for sending this information to the nearest hospital.
Forecasting COVID-19 cases :
We developed a model using Prophet that can forecast future covid-19 cases, deaths, recoveries, and active cases for the next 7 days. We did this for 12 states of India. And after that, we used the folium library to generate a map and visualize the results in the form of a hotspot on a map for each state. We used the open-source time-series dataset provided by the John Hopkins university
We also created a dashboard using the 3rd party software called amcharts that visualizes the world-wide covid-19 cases, deaths, recoveries and active cases for each day.
Experience and future work :
It was a great experience for us working on a real-world problem and we learned new technologies while working on these modules. We faced a lot of challenges, one of the biggest challenges is how to count the no. of people when a group of people is overlapped. And when it comes to forecasting, in the future we want to forecast covid-19 cases for all the states and also districts within the state. And also use hybrid machine learning techniques like ensemble learning to improve the accuracy of the models.
Acknowledgement:
We would like to thank Dr. Kuldeep and Dr. Suneet Gupta for guiding us through the development of the project .
Thank you for reading my blog.
Baipureddy Neeraj, VVS Manideep
B.Tech 4th year CSE,
Bennett University.