The 20-person company is called Pano AI, short for the panoramic cameras it has been installing on Northern California mountaintops since its launch last summer and the artificial intelligence software that powers its system. A rotating camera snaps 10 photos of the surrounding landscape every 60 seconds and stitches them into a time-lapse stream of high-definition images that are then scoured for wisps of smoke by machine-learning software.
When Pano's AI flags a possible fire start, an employee receives an alert to verify it visually. If it appears to be a genuine fire, Pano notifies firefighters with the exact coordinates of the blaze.
"We want to maximize the information available to first responders in the early moments of a fire," said Pano CEO Sonia Kastner. "The idea is to nip these in the bud before they turn into unstoppable infernos."
Kastner founded Pano after 12 years in manufacturing and product development for Bay Area tech companies including Nest, the Google-owned venture that uses video cameras and artificial intelligence to help people monitor their homes. With wildfires an increasing threat to Californians, Kastner researched the needs of firefighters and "heard a chorus of calls for more technology solutions," she said.
As part of its pilot program, the company has engaged local fire districts and utility providers and installed 23 cameras across four states — California, Colorado, Oregon and Montana. Of those installations, 14 are located in the Bay Area, ranging from the Mayacamas Mountains in Wine Country south to the Santa Cruz Mountains.
Early detection and rapid response are the strategic cornerstones of wildfire suppression. Traditionally, forest rangers kept watch over vast, wooded landscapes from the high perches of fire towers and radioed authorities when they spotted smoke plumes above the treetops. But deploying digital eyeballs that monitor remote areas 24/7, sometimes in thermal infrared, can shave precious minutes off response times.
"Even if you get an extra two minutes, and you can triangulate where the fire is, that's huge," said Jim Comisky, a retired fire chief who is president of the South Lake County Fire Protection District Board of Directors.
South Lake County has engaged Pano to beta test four camera installations in the peaks around Middletown, including on Mt. St. Helena. The new technology has been net neutral in helping with early detection, Comisky said — about as quick and accurate as 911 calls from people on the ground.
If the network expands, Comisky believes there could be meaningful benefits. Kastner estimates that 1,000-2,000 strategically placed Pano cameras would be needed to cover all of California.
Using webcams to spot wildfires isn't a new idea.
Academics in California, Nevada and Oregon pioneered a wildfire monitoring system that took root eight years ago in the Lake Tahoe basin. The initiative, called ALERT Wildfire, has since expanded to about 975 cameras, including 850 in California. It functions as a nonprofit and installs video cameras at ski resorts, on Internet towers and in open spaces in partnership with private citizens, federal land managers and utilities like Pacific Gas and Electric.
Unlike Pano, which charges subscription fees to access its feeds, the ALERT streams are publicly available. However, not all of ALERT's cameras are in constant 360-degree rotation and require human observers to spot fires via the feeds.
"We have people in New Zealand helping spot fires in California," said Graham Kent, director of the Nevada Seismological Laboratory and chief architect of ALERT Wildfire.
That's where the AI proposition becomes promising: no humans are required in the earliest phase of detection.
ALERT recently partnered with a Korean company called Alchera, which is running AI on more than 100 of ALERT's wildfire cameras. Earlier this year, Alchera entered into agreements with PG&E, Sonoma County, San Diego Gas and Electric and NV Energy to provide wildfire monitoring and alerts.
"You can see companies starting to flow into this space because there's a need to do better," Kent said. A future collaboration between ALERT and Pano isn't out of the question, he added. "Eventually, the techonology that is most successful in detection will be on the widest camera network."
One drawback of relying on early-stage AI is the potential for high volumes of false-positive results — alerts about, say, a cloud formation the software mistakes for a smoke plume. However, Kastner says Pano's system — including its human verifier — identifies fires with a 90% or higher rate of accuracy.
Ideally, the system would eventually function like a home carbon-monoxide detector, quietly checking for danger at all times and blasting an alert when smoke arises — but for entire states. That's the idea that excites fire authorities like Comisky.
"The fire service has changed," Comisky said. "If we can utilize some of this new technology, it could be game-changing."
©2021 the San Francisco Chronicle, Distributed by Tribune Content Agency, LLC.
