How Computer Vision Can Help Fight the Pandemic

The coronavirus outbreak sparked interest in computer vision applications in healthcare. The pandemic prompted governments to put out new health-related policies and regulations that require constant surveillance of citizens. But, monitoring large areas without the help of technology is next to impossible. The healthcare industry is already under a lot of pressure from this unprecedented situation.

The best way to stay safe during the pandemic is to wear a mask and keep the social distance. People with well-known COVID-19 symptoms must self-isolate. Higher body temperature is the most notable COVID-19 symptom, and most checks focus on it. Computer vision applications can aid the government in fighting the pandemic by enforcing the rules necessary to curb the spread of the virus.

The idea is to use regular and thermal cameras for monitoring. Real-time image processing algorithms try to locate anyone who is breaking the rules. Data scientists might approach the problem better than others.

Computer Vision Applications for Thermal Detection

Hospitals, airports, and train stations are the most dangerous places during the pandemic. Staff take people’s temperatures before they enter trains and planes, or hospitals. The solution is far from perfect. It creates crowds and exposes staff to potential infections.

Data scientists thought of using computer vision to track body temperatures. The field of computer vision combines artificial intelligence with thermal image processing. Only when someone crosses the upper limit of 37.5°C (99.5°F), the staff can react and deny entry to them.

Thermal detection using computer vision

Credits: Thermal imaging: Learn the limits of elevated body temperature screening - Vision Systems Design

How Does Thermal Detection Work?

Even though it sounds simple, thermal detection is a complex machine learning process. Artificial intelligence models do image processing on low-quality, moving images. Here are some problems for data scientists to consider:

  1. Thermal cameras measure the infrared radiation from the first surface it sees.

Pointing at a human being, you can only measure the temperature of the skin. However, most of the time face skin temperature does not correlate to the actual body temperature.

Scientists have shown that the inner canthus is the best representative of the actual temperature. The inner canthus is between 5 and 7 mm in size. The area needs to be covered with as many pixels as possible for the best result. Still, results will vary from person to person.

Thermal Detection Inner Canthus

Credits: Thermal imaging: Learn the limits of elevated body temperature screening - Vision Systems Design

  1. Thermal cameras have low resolutions.

The typical thermal camera resolution is 320 x 340 or 640 x 480. It has the greatest detection distance and the number of people it can scan at the same time.

  1. Thermal detection is often not accurate enough

The typical absolute accuracy of thermal cameras is +/- 0.5°C. For other applications, it might be good enough. Still, for human body temperature, it is the difference between healthy and sick.

The data scientist needs to take these factors into account. Developing machine learning algorithms or neural networks for these models is no joke.

Computer Vision for Social Distancing and Mask Detection

Computer Vision for Social Distancing

Deep learning techniques for person tracking can help fight the pandemic. A successful deep learning model only needs to calculate the distance between people. The combination of computer vision and social distancing looks like this:

Social Distancing with Computer Vision

Credits: Using AI to Detect Social Distancing Violations | by Priya Dwivedi | The Startup

Some person tracking deep learning algorithms are available online as open-source code. For your applications, you might want to hire data scientists to improve the results. Here are some issues to consider:

  • The camera image is 2D

If you want to calculate distance precisely, you will need to know the camera's position relative to the ground and make in-depth calculations.

  • Person identification issues

The machine learning algorithm should not count the same person twice. Also, it mustn't falsely identify a person.

  • Surveillance network problems

If there are many cameras connected in a network, following someone across many cameras can be challenging. Moreover, the hardware required to run the artificial intelligence algorithm gets more expensive.

Computer Vision For Mask Detection

Recognizing who is not wearing a mask is a relatively simple data science problem. Making it work with low-resolution cameras is challenging. So we recommend you to avoid open-source machine learning solutions and hire a data science company instead.

Essentially, there are two main tasks in mask detection:

  1. Object detection with a neural network

In this step, object detection refers to facial recognition. It is intriguing to see how facial recognition works with face masks.

Classification of faces (with/without a mask), also with a neural network

This step also uses object detection, but this time, the object is the face mask.

Even the most basic methods will work well with close-up shots. But, image resolution and weird angles from surveillance footage can make the process more challenging. Only the most experienced data scientists know how to approach such problems. Here are some examples of surveillance photos the machine learning algorithm needs to classify:

Mask detection with computer vision

Credits: Face mask detection in street camera video streams using AI: behind the curtain - Tryo lab

Object detection is easy with high-definition portrait photos. Surveillance photos are far from ideal for object detection and facial recognition. But, that is where you can recognize the proper data scientist.

Computer Vision for Contactless Access Control and Identification

Minimizing contacts with other people is driving the development of biometric technologies. Contactless access control and identification systems should replace security workers and porters. These systems mostly rely on computer vision and facial recognition.

Such systems are often combined with mask detection and social distance detection tools. This way, large companies can control access and identification. Facial recognition helps employers track their employee's behavior during the pandemic.

Computer Vision and Deep Learning for Radiological Image Analysis

Healthcare computer vision and deep learning methods can help with X-ray and radiological imagery. First, the pandemic leads to an increase in X-ray scans. It is a growing issue in the healthcare industry. When a machine learning algorithm classifies the scans, it can help doctors focus their time on other challenging tasks.

However, there is a more interesting aspect of computer vision for COVID-19 applications. There have been many data science approaches in using deep learning for detecting COVID-19 from X-ray scans.

At Cranfield University, students have developed fascinating deep learning models. The models help students analyze X-rays and diagnose them for COVID-19. It is based on pneumonia as one of the significant COVID-19 symptoms. The first classification determines which scans are positive for pneumonia. The second classification tries to diagnose for COVID-19.

X-ray of pleumonia computer vision

Credits: Application of Computer vision in Health care | by Anushka Srivastava - Mediu

Future data scientists from Cranfield University claimed that their approach resulted in accurate diagnosis. They used regular machine learning algorithms combined with deep learning models. Despite initial good results, they intend to continue their research. They want to come up with more complex and more precise algorithms.

Dr. Zeeshan Rana, the leader of the data science team behind the research and a lecturer at Cranfield University, said: "The research carried out in this pilot project has led to some extremely promising results, and we are looking to build on this success rapidly to help in the fight against COVID-19. I am incredibly proud of the work my researchers have carried out. They are a credit to the University and I'm delighted that we are able to support them remotely in carrying out their studies."

Final Words

Unprecedented times call for radical measures. Governments look to find solutions to make people take the situation seriously. Governments and large corporations need an efficient way to enforce pandemic-related guidelines. Data scientists might have a solution.

As always, artificial intelligence saves the day. Combining computer vision applications with machine learning algorithms leads to real-time monitoring models. The models can point out people who are not following the rules and alert the staff. Also, computer vision systems can be used for identification and access control in large companies. The combination of the two is a well-rounded solution for most companies. However, it is crucial to hire the right data scientists for the job!

Finally, the field of computer vision brings new ways of detecting the COVID-19. With large enough datasets, new patterns will emerge, and we will find out more about the virus. Healthcare computer vision systems might change healthcare as we know it. Some things are invisible to the human eye. Yet, the artificial intelligence algorithm sees it as a non-disputable pattern. The data science behind it may be tricky, but it can make our world a better place. Who would have thought to use computer vision for fighting the COVID-19.


Michael Yurushkin

Founder of BroutonLab, PhD