Methods and evaluation of a robust emergency vehicle detector with acoustic sensors in context of autonomous driving for German siren signals

Abstract:
Autonomous driving vehicles require reliable detection of emergency vehicles to provide them a clear path in case of an emergency. This paper provides a workflow for monitoring the surrounding with respect to specific events and transferring this into valuable information using microphones and the state-of-the-art low latency A2B transmission bus. For the evaluation of the proposed emergency vehicle detection stage, a method for collecting data and setting a reference on the collected data is given. As the process of collecting data is time- and resource-consuming, we present an easy-to-use method of simulating acoustic driving scenarios in the context of emergency vehicle detection. The conformity of simu-lated and real-world data is shown as high. The emergency vehicle detection system in combination with an exterior microphone delivers a reaction time on an approaching emergency vehicle in a motorway scenario comparable to that of a human driver. This fulfils the current legal demands. Furthermore, a high robustness towards a variety of other environmental sound events is shown.