When sensors listen «around the corner»
You can't always see drones - but you can almost always hear them. An acoustic detection technology developed by the Fraunhofer IDMT is designed to detect unmanned flying objects even outside the field of vision. This could be a crucial addition to existing security concepts.
Drones have long been part of everyday life - as a leisure device, as a tool in industry and the media, but increasingly also as a risk in security-critical areas. Because as soon as an unmanned flying object approaches sensitive zones without permission, things can quickly get tricky: at major events, at airports or around critical infrastructure. The main problem here is that conventional detection methods do not work reliably in every situation. Optical systems rely on a clear view, radar can be restricted by terrain or vegetation, and radio reconnaissance reaches its limits when drones transmit no radio signatures at all or only radio signatures that are difficult to detect - for example in the case of fiber optic or fiber optic-controlled systems.
Supplement to radar
This is precisely where the solution from the Fraunhofer Institute for Digital Media Technology IDMT in Oldenburg comes in: an integrated acoustic sensor solution that detects and localizes drones even outside the line of sight. «The approach is expressly intended as a supplement to radar, cameras or lidar - and closes a gap,» explains Christian Rollwage, who is leading the project. «Acoustics do not require a line of sight and can therefore also provide information where other sensors are »blind', for example in wooded areas or urban canyons." The system listens around corners, so to speak, and can therefore detect drone operations even in built-up or wooded areas.
Machine learning instead of continuous monitoring
The core of the technology is the automatic classification and evaluation of sounds. Paul Reuter, project team member, describes the technical basis as a machine learning process, specifically deep learning methods with neural networks. «Spectrograms - i.e. representations of the frequency content of an audio signal over time - often serve as the basis.» Characteristic patterns emerged in these sound images: «Drones generate a fundamental frequency and overlying harmonics through their rotors, which form a typical harmonic pattern. The system learns this pattern during training and can later decide in real time: Drone - yes or no,» says Christian Rollwage.
«Acoustic drone detection is not a replacement for existing systems, but a useful addition where radar, cameras or radio reach their physical limits.»
From the user's point of view, the decisive factor is that the system is designed for automation. It should not have to be permanently monitored by personnel, but should trigger a message or alarm chain when a drone is detected. Reuter points out that forwarding to control center or customer systems is also planned in principle - for example as an edge/IoT approach, in which evaluation and reporting take place as close as possible to the sensor.
Range, limits and robustness
As with all acoustic detection, the range depends on the conditions - especially the signal-to-noise ratio. «Wind and ambient noise, such as traffic, can make detection more difficult. Under favorable conditions, ranges of around 100 to 200 meters are possible,» says Rollwage. At the same time, the researchers are working on reducing interference at the hardware level. Reuter describes wind as a particularly critical factor: «Strong turbulence directly at the microphone can mechanically superimpose the signals - which is why wind protection is part of the system design.»
Another advantage: compared to high-resolution radar or camera systems, the acoustic solution is low-energy and therefore suitable for autonomous battery operation. A wake-up concept is also possible: after an initial acoustic contact, further sensors can be activated in a targeted manner - an approach that saves resources and increases overall safety at the same time.
Sensor data fusion: fewer false alarms, more safety
In practice, acoustics should not stand alone, but be embedded in a sensor data fusion. This combination increases robustness because several modalities can verify the same incident. Rollwage points out that false alarms are particularly problematic: «If a system strikes too often without a real incident, attention drops - and in an emergency, the response is too late.»
According to the experts, the technology can achieve 360° coverage. An extension to other acoustic events, from vehicles to the sound of gunshots, is conceivable. Thanks to their good availability, the acoustic sensors can be deployed across the entire area.
Fields of application: from events to critical infrastructure
Possible areas of application for the technology could include the protection of prisons, major events and properties. Airports are also considered particularly sensitive because drones can cause considerable damage there through collisions or interference. «This also includes the protection of test sites, where new vehicles, known as prototypes, are tested, for example,» says Reuter.
Data protection and product maturity
Data protection also plays a role in legal terms - especially because microphones are considered sensitive in many contexts. According to Rollwage, however, the system is designed in such a way that no audio data is stored permanently: The analysis takes place on the sensor or locally, with only a very short, fleeting storage window of a few seconds. In addition, algorithms could - if necessary - recognize voices and treat them accordingly.
The system is not yet a finished end customer product in the traditional sense. The IDMT is developing the technology to a high level of maturity and is then looking for partners to bring it to product maturity. Accordingly, the offer is primarily aimed at companies from the defense and security sectors, manufacturers of existing drone detection systems and system integrators who want to integrate acoustic sensors as an additional detection layer.
