While some vehicles now utilize computer vision systems to "see" items such as road signs and pedestrians, the urban environment consists not only of sights, but also sounds. With that in mind, German scientists are creating a noise-recognition system for cars.
Currently being developed by a team at the Fraunhofer Institute for Digital Media Technology, the prototype setup utilizes of an array of microphones that are housed within an aerodynamic fin that's mounted on the vehicle's roof – it looks much like a modern antenna.
Audio captured by those mics is relayed to a separate control unit. It, in turn, contains a microprocessor running custom software based on artificial intelligence-based algorithms. Those algorithms were initially "trained" using a database of prerecorded sounds that are considered important for drivers to hear, such as the approaching sirens of emergency response vehicles, the shrieks of children playing, or even the clicking of a nail in a tire.
After suppressing distracting background noise and boosting the acoustic signatures of more distinctive sounds, the software is thus able to identify things such as sirens. In fact, by analyzing the changing levels at which a siren is picked up by the multiple microphones, the software can determine the direction from which the fire engine, etc is approaching.
Drivers would subsequently be alerted to the situation, so they could take appropriate action. In the case of self-driving cars, the vehicle might even automatically pull itself over to the side of the road.
The technology is expected to be commercially available "by the middle of the coming decade." It could also have applications in other fields, such as listening for malfunctioning machinery in factories.
An already available system, Cerence EVD, uses newer cars' existing in-cabin microphones to detect sirens. Google spinoff company Waymo has also developed a system of its own.
Source: Fraunhofer