By: Christina Lee (CC '21)
I saw that past projects have included devices that improve pedestrian safety. Can you first speak a bit about your background and what types of projects your lab has done in the past?
My background is in wireless embedded systems. Before joining Columbia, I was at Microsoft Research and then Intel Labs. My research has been focused on using embedded platforms, such as the smartphone, combined with data analytics methods, such as machine learning, to solve real-life problems. I think that’s something that has carried over with me when I joined Columbia five years ago. My lab is trying to work on research that tackles real-life problems. For example, in the pedestrian safety project, we developed technology that could enable pedestrians to better sense the world around them. They can be alerted in advance when a vehicle is on a collision trajectory. A lot of our other projects also have similar real-world motivations.
I read recently that you have been developing a way to screen large groups of people for fevers. Would you be able to speak a bit more about this project and how it came about?
This project is something we have been working on before COVID-19 started. The project was, at the time, not focused on fever detection, but instead on thermal comfort at scale. It was a project on how to improve energy consumption in buildings while optimizing individual occupant comfort. An important component of that project was to understand whether people are comfortable or not, and that’s actually quite difficult at that scale. One could write an app where the user tells you whether they are comfortable, but that requires direct input from the user. So we wanted to develop technology that could measure the skin temperature of every occupant in the building, which could easily adapt to fever screening. So that’s how this project came about. When the COVID-19 situation started, the hospital that my wife works in was facing a lot of shortages in PPE and personnel. During the check-in process, they needed to take the temperatures of patients. Since they had a lot of patients coming in, the check-in process was very slow, so they needed a better solution to screen patients. This project would potentially address this need.
How has your progress been, and have you encountered any setbacks?
So there are some technical challenges we are addressing right now. Simply relying on an inexpensive thermal imager that costs $300 is not accurate. You would need something in the order of $25,000 to get an accurate thermal image of a large space. We are trying to find a way to improve the accuracy of thermal images taken using a low-cost thermal imager. We developed a data-processing pipeline where we can first create a 3D model of every single person in the frame. Once we have a 3D model of every person, including his or her head, we can then map the images that we’ve taken so that instead of relying on a single image, we can add more images to the 3D model of the human head with the correct orientations. Since a person is not always looking at the thermal imager, the traditional approach is not going to get a very consistent and accurate shot of the face. With our approach, regardless of which side of the face is facing the camera, we can conform it to the 3D model of the head so we can continuously improve screening accuracy with the data that is incoming in.
There is also the other question of calibrating the camera because different distances, humidities, and temperatures can all affect how pixel values translate to actual skin temperatures. This is something that requires time and many experiments to obtain calibration data, so that’s what we’re working on now. We have a working prototype, but we’re still trying to improve it.
What do you think the public should know about the timeline of your research? What is your perspective on the future of your research and its impact on COVID-19?
We’re trying to publish the design and the software within the next few months. Since this is something that would be incredibly useful for the public, we have been working on an accelerated time frame. We want to make sure that we can get our system out to the public within the next few months. We are hoping that once we have hospitals, schools, and other places use our system, we are able to receive feedback, build a community around this idea, and get others to help improve this system.
In terms of a longer time frame, this is something that is useful beyond COVID-19. Creating this system at a low cost is useful for any kind of illness where fever is a symptom. So I really hope that this will have a long-lasting impact on the healthcare industry and many other places that can use a system that provides continuous detection of fever within a population. So maybe in the future, schools, cruise ships, buses, or subways can use this, and I really feel like this could be a game-changer.