San Francisco – An Israeli start-up believes it can solve some of the tricky problems faced by self-driving cars by employing thermal cameras to detect heat from pedestrians, animals and objects, the latest technology being tested in the fast-growing industry.
Start-up AdaSky announced on Monday that it had developed a far infrared (FIR) thermal camera that works with computer vision algorithms to detect people or objects on the road. The camera is designed to help in so-called “edge cases” in autonomous driving, where other sensors might fail.
Automakers and their suppliers working on self-driving cars have been experimenting in recent years with forward-facing cameras, radar, and Lidar, which uses light pulses to “see” the road. Each has its strengths and weaknesses, and all elements must communicate with each other.
“The (automakers) are now beginning to understand it’s not enough,” Eyal Madar, AdaSky’s chief technology officer, told Reuters. “They need a new sensor … to help them with edge cases.”
Thermal cameras are a mature technology initially developed for military use. Madar said AdaSky is focused purely on applications for autonomous vehicles, unlike much larger competitor FLIR Systems, which is more broadly focused.
“In the last four to five years the technology has become more available and cheaper,” Madar said, adding that AdaSky’s price at mass production could be a few hundred dollars. “Thanks to that, it has become possible to think about the camera for autonomous driving.”
Global automakers and top suppliers were validating AdaSky’s system, said Madar. He declined to name the companies.
A full self-driving prototype of AdaSky’s system would be ready within a year, with mass production two to three years away, he said.
The company says a thermal camera can outperform a traditional camera at challenging times such as at night, in fog or rain, or at sunset or sunrise.
Another difficulty for self-driving cars are their inability to differentiate between images and the real thing. For example, the image of a bear in an advertisement plastered across the back of a truck may be viewed by the car following the truck as a real bear, causing the car to brake.
In detecting the differences in temperatures emitted by living objects and inanimate objects, however, the thermal camera would help avoid this scenario.
AdaSky’s technology can determine whether an object is a pedestrian at 200 meters (219 yards), and from further away for larger objects like cars, Madar said.