There’s that iconic moment in many a Superman story in which the Man of Steel is soaring through the sky when, thanks to his super hearing, he picks up the faint, frightened voice of a person calling for help. Without a second thought, Supes zeroes in on the audio source and flies off in that direction, ready to perform a daring rescue.
Swap out comics’ oldest cape-sporting superhero for a drone and you’ve pretty much got the technology being developed by researchers in Germany’s Fraunhofer FKIE research lab. They have kitted out drones with microphone arrays, some smart artificial intelligence algorithms, and an onboard computer in order that it can recognize the sound of screams and detect their precise location. At present, the team is using DJI Matrice M600 drones for the proof-of-concept task, although these UAVs could easily be swapped out for just about any drones that are able to carry the requisite 2.2-pound (1-kilogram) payload.
“We are developing technology that could definitely save lives when disaster strikes,” Macarena Varela, a research fellow at Fraunhofer FKIE who worked on the project with Wulf-Dieter Wirth, told Digital Trends.
Every minute counts
Varela points out that, after a disaster such as an earthquake, every minute of the rescue effort is crucial. With victims potentially trapped in rubble or otherwise injured, every additional 60 seconds could very well make all the difference between life and death. Drones are already used in disaster scenarios due to their ability to survey scenes from the air, combined with their agility and ability to access difficult-to-reach areas. But until now, drones have largely had to rely on visual identification to spot people. However, using the so-called “Crow’s Nest Array” of acoustic technology, they could soon be able to fly to a disaster zone and then listen to identify people by their voices as well.
In an age of smart home assistants like Amazon’s Echo, the idea of A.I. tech that can detect voices isn’t quite the science fiction it once was. But while Alexa or Siri may be able to make out your voice from across the kitchen, with food bubbling on the stove and music playing in the background, that’s nothing compared to the challenge the team’s scream-sensing drones face here. No regular household din compares to the wind and rotor sounds, plus other assorted environmental noises, that this drone setup has to contend with — and manages to navigate with aplomb.
“Currently we are still processing the data to appropriately filter those noises and achieve a better reading on impulsive sound events,” said Varela. “Over the years, we have gathered extensive experience in filtering noise to detect different types of impulsive sound events. For instance, in the past we successfully suppressed extremely noisy helicopter sound to detect impulsive sound events. We also filtered ground vehicle noise for the same purpose. By applying appropriate filtering methods, we achieved robust detection of signals of interest.”
When Varela refers to the past, she’s not kidding: This project has been the focus of (to date) eight years of research at Fraunhofer FKIE. Previously, the team demonstrated that they were able to detect and localize sounds such as screams in noisy environments. However, this required a much larger and heavier microphone and processor array that couldn’t possibly have been carried by a drone. By comparison, this latest demonstration was carried out using digital MEMS microphones, the same small and lightweight microphones found in cell phones.
More work to be done
Don’t think that this is the end of the project, though. “We are still developing the system: Adapting the methods, working on noise suppression methods, geographical localization estimations, and more,” Varela said. “We would like to develop it further. [One goal is that] the drone must be able to automatically fly [to] areas of interest.”
They also aim to attach other sensors to the drone setup as a way to be able to autonomously provide information about victims to emergency crews.
Factor in the other emergency rescue robots and drone technologies currently being explored — ranging from underground rescue vehicles to fire-extinguishing robots — and it’s not impossible to think that the rescue missions of the future could be carried out predominantly through automated technologies. That may still be some time off, but work such as this demonstrates just how much closer it’s getting.
This scream-sensing research was recently demonstrated virtually during the 180th Meeting of the Acoustical Society of America, held from June 8 through June 10.
