Startup Uses AI, thermal imaging and SaaS to Streamline Process of Detecting Fevers

In the wake of the coronavirus crisis, Buffalo Automation is piloting a new project that is adapting its technology to streamline the process of taking body temperatures in the workplace.

The company is running a pilot program, which is called the Bifrost project, at a facility operated by UB Neurosurgery (UBNS), a member of the UBMD Physicians’ Group, the practice plan of the Jacobs School of Medicine and Biomedical Sciences.

Bifrost equipment — which uses thermal imaging to measure people’s temperatures — was installed this summer at the UBNS Clinical Neuroscience Center, located on George Karl Boulevard in Williamsville. The equipment scans people as they walk past a revolving door from an indoor balcony. Smaller, battery-powered units were also installed in two check-in kiosks.

Currently, the equipment is gathering data, validating the ability to do long-range facial detection and feature extraction using thermal images. In the next phase of the project at UBNS, it will begin to compare and validate temperature differences between facial features using an instrument called a microbolometer, which detects infrared radiation.

“As we see many patients in the community, Buffalo Automation reached out to explore synergies that may exist,” says Elad I. Levy, MD, SUNY Distinguished Professor and L. Nelson Hopkins III, MD, Professor and Chair of neurosurgery. “I think as researchers at UB, we all share a common interest in using science to augment health in our community and globally.”

According to Vikram, Bifrost got its start early in the pandemic, when shipping companies that the startup had contracted with were reluctant to proceed because they feared Buffalo Automation staff would inadvertently infect ships’ crews, in spite of the “extreme precautions” the Buffalo team was taking.

“One of us then flippantly suggested in the office that AutoMate should be able to tell if our engineers had a fever before embarking on a vessel … and Bifrost was born,” Vikram says.

“Bifrost does not need super-expensive and periodically calibrated thermal cameras to work,” he says. “It is essentially a simple software- as-a-service solution that makes an existing thermal-camera solution work just right and, most importantly, is a viable solution for scanning large crowds from a distance.”

Vikram describes Bifrost as a convolutional neural network that identifies and extracts the inner canthus — the inner corner of the eye — from each human face contained in a given thermal image.

It then extracts other facial features, such as foreheads, noses, ears and necks, from each human present in an image. After comparing patterns in the metrics in various individuals, it isolates individuals with abnormal temperatures.

In other words, Bifrost takes specific human features from thermal images and measures the skin temperatures from areas most likely to indicate a fever.

“The ultimate goal of the pilot project is to prove that Bifrost is a better option than using a handheld infrared measuring device by comparing rates of false positives and negatives,” Vikram says.

If a fever is detected, there will be a secondary screening using traditional methods for any individuals flagged by Bifrost, he adds.

The anonymized information will not be shared with any outside entities, Vikram says. However, images will be sent to Buffalo Automation for training purposes.

By taking the temperatures of multiple people simultaneously, Bifrost hopes to eventually streamline a tedious and expensive process that requires trained personnel and often results in people having to wait in line — thus saving time and money.

While some may think the technology could present privacy issues, Vikram maintains it is conceptually similar to closed-circuit television cameras being used in public areas for security reasons.

In fact, he says, “Bifrost offers more privacy, since it is very difficult to recognize someone’s face in a thermal image.”