The cheapest are temperature guns being handed to security guards to point at visitors. Then there are thermal cameras being placed alongside the usual security cameras at entrances. These cameras, mostly sourced from China, are of dubious virtue apart from their low cost.
IPVM, an independent site monitoring video surveillance, recently reported that many thermal camera manufacturers had made false claims about accuracy. Some were palming off heat-sensing devices meant for detecting fire or the presence of humans as thermal scanners for covid. So the question is how to make the cameras effective while allowing for rapid deployment at scale.
Bengaluru-based startup Tonbo Imaging, which provides night vision and other equipment to modernize defence forces around the world, has an answer. It has adapted its technology to febrile temperature screening.
One issue with most cameras is the low resolution of sensors. They can’t distinguish facial features from a distance of 5-10 metres. Getting a reading from a random part of the face isn’t accurate—key areas are the forehead and corners of the eyes.
Tonbo works at scale with contract manufacturers for higher resolution thermal cameras in military applications. Now it is repurposing those into temperature screening cameras, which are being deployed around the world.
“My team tested nearly ten systems from US, Germany and China," says Dr Valerie Solan-Gutarrez, CMO, Medecins Sans Frontieres, who is overseeing materials and technology for screening and assessment for Latin America. She is overseeing deployment of Tonbo’s systems in Brazil, Peru and Chile.
“We chose Tonbo because its cameras had double the resolution of others we tested. They gave us field accuracy of 0.25 degrees Celsius while other systems were plus/minus 3 degrees Celsius. A programme supported by WHO will use hundreds of their systems across South and Central America."
A key piece of the solution is artificial intelligence. Tonbo’s partner is Nvidia, a global leader in AI systems, which launched an application framework for smart hospitals called Clara Guardian last month. Nvidia’s hardware for AI enables real-time video analytics for automated processing. This is needed at airports, hospitals and other places where a person with a laptop can’t monitor large numbers of people.
It’s not just their temperature being measured. The cameras also spot people in a crowd who are sneezing or coughing, and whether they’re wearing masks and following social distancing norms. Facial recognition technology kicks in to flag those who may be likely to spread infection.
A lot of Tonbo’s camera systems for military use are also built on Nvidia’s platform—so it was more of an extension to integrate its covid-related sensors and algorithms with the Nvidia hardware.
“Not much needs to be done to adapt our defence equipment to medical use," says Tonbo founder and CEO Arvind Lakshmikumar. “We’re telling our customers in India, the Philippines and other countries where we’ve sold a large number of thermal imagers that they can repurpose them because there’s a paucity of equipment for accurate temperature screening. For that, we’re giving them an additional AI processing box connected to what they already have."
One reason Tonbo faces little resistance in doing this adaptation is that it focused from the outset on designing its defence equipment with off-the-shelf consumer electronics components. Its value proposition came from real-time AI analytics on images captured from multiple sensors.
Its aim was to make modern systems available to defence forces around the world—and not just the leading ones like the US special forces. That approach also applies to the thermal imagers it is repurposing for covid screening.
CONSUMER ELECTRONICS BASE
“Our camera systems have been based on consumer electronic hardware, such as Qualcomm and Nvidia chipsets. They’re made at contract manufacturing facilities in large volumes. It means we’re building them at a price point that’s as commercial as it gets," says Lakshmikumar.
“They’re cheaper than similar systems from Germany but more expensive compared to many systems that come from China. That’s because Tonbo’s systems are more accurate," says Dr Solan-Gutarrez. “We care about accuracy. There’s no point in having a fever scanning camera just for the sake of it."
The first large scale users in India will be all the military bases which have large numbers of people.
“Currently the screening cameras being fielded in India are either from China or Chinese knockoffs. They don’t perform terribly well. Besides, the military establishment will not want Chinese equipment in critical areas where our security is involved," says Lakshmikumar.
“The Northern Command was one of our big customers. They have thousands of units in Jammu. So, wherever our cameras are already deployed, they will be the first to start using the systems for temperature screening."
Military hospitals followed by civil hospitals will be the next target group for the Tonbo systems meant for sophisticated screening. Then large tech company campuses in Bengaluru are likely users.
“We’re also integrating it with iris recognition systems which automatically plug into the attendance management register," says Lakshmikumar. “So, when you enter a building, the iris system will recognize you, measure your temperature, see if you’re wearing a mask or other protective gear, and log you in directly. If you don’t meet all the parameters, it will raise an alarm and you will need to go and fix the problem before you’re let in."
What’s needed today goes way beyond biometric or other IDs. All the safety norms for covid-19 have to be added on. That’s very hard to do with manual checking in places with large workforces. Hence the need for an automated system.
“When large volumes of people come back to work, you need a system that automatically checks everything with a high level of accuracy," says Lakshmikumar. In the case of Tonbo’s AI-enabled thermal cameras, the margin of error is a mere 0.1 to 0.3 degrees Celsius.
Malavika Velayanikal is a Consulting Editor with Mint. She tweets @vmalu.