When it comes to autonomous driving, people usually talk about what the car can or cannot see. It sees the pedestrians, it sees the signs and sees the other cars on the road. Sensors perceive the entire environment, but this perception goes far beyond “seeing”. In the future, as current research by the Fraunhofer Institute for Digital Media Technology IDMT shows, autonomous vehicles will also have a sense of hearing. Researchers in Oldenburg have developed a first prototype for recognizing external sounds, such as a siren.
We know it from our own way of dealing with traffic. Unless we are on the phone or have headphones in our ears, we often let our hearing be our guide in our everyday life on the road. We hear from which direction the cars are coming, hear ambulances or police. So if the car is to be able to perform all the activities that we currently perform so intuitively behind the wheel autonomously, then it needs a sense of hearing. According to information from the Fraunhofer Institute, this sense of hearing consists of sensors and systems that are able to perceive and interpret external sounds. In combination with the radar and camera sensor technology already installed, this is a real added value for the autonomous future.
“For autonomous vehicles, external acoustic perception systems do not exist yet, despite their high application potential. They signal, for example, in a fraction of a second when a vehicle with the siren switched on is approaching. In this way, the autonomous vehicle knows that it must swerve to form an escape route” – Danilo Hollosi, Group Leader Acoustic Event Detection, Fraunhofer IDMT in Oldenburg
Noise analysis using artificial intelligence
The team of the research institute, which comes from different disciplines, saw the optimal signal recording as a particular challenge. In spite of all this, the car must be able to react to the slightest noise in road noise. The sensors therefore not only have to process the acoustic signals, they must also rid them of background noise and locate their origin as precisely as possible.
To do this, the algorithm behind the sensors was trained using machine learning with the acoustic signatures of relevant sounds, such as the siren or a horn. The result is a database that all vehicles can access and which is constantly updated during operation.
Together with its partners, the Fraunhofer wants to realize the first prototype and make it ready for the market by 2025. According to the researchers, the new technology can also bring remarkable benefits for other industries. For example, in quality assurance in industrial production.
Post picture: Fraunhofer Institute