This project investigated the feasibility of a system for acoustic tunnel monitoring. The sounds typical in the operation of tunnels are characterised by the engine, rolling and flow sounds of the passing vehicles. Any anomalies in the sound, such as e.g. a vehicle crashing into the tunnel wall, two vehicles crashing into each other, squealing tyres, voices (shouts), etc., as well as anomalies in the sound of individual vehicles can be detected by microphones placed in the tunnel.

With special detection algorithms, it is possible to identify these sounds automatically, and to allocate them to specific alarm classes. These sounds occur right at the time of the incident - not after some time has passed - and can thus be detected. If an incident is detected in a certain section, various measures can be activated immediately. For example, an acoustic signal depending on the alarm or incident class can sound in the tunnel control centre, and the camera image from video monitoring of the relevant section can be switched to one central monitor. This allows the tunnel manager to assess the situation immediately and take the necessary actions. Thus, valuable time is gained and maximum emergency aid and accident prevention can be provided to both the people involved in the incident and to following vehicles.

Starting out from a study of the room acoustics of tunnels, which have a strong influence on the spreading of sound, a recording system was developed that allows continuous recording of the sounds in a tunnel for several weeks. These recordings were necessary in order to obtain the test data required for development of the classification algorithms. These test data include both “normal” operating sounds and “alarming” sounds that were recorded in various tunnels with various operating modes and tunnel geometries in the course of the project. The sound database thus obtained serves the development and prototype implementation of the classification algorithms, which form the core of the future system for acoustic tunnel monitoring. To illustrate the function of the algorithms, demonstration software has been implemented.
In the course of development of the algorithms, 15 characteristic properties were found based on which the sounds can be classified.