Analysis of microphone use for perception of autonomous vehicles

Patryk Siwek

Abstract: Many teams focus on the implementation of vehicle perception with the help of artificial intelligence using data from cameras, lidars or radars. A less common approach is to use sound sensors like microphone arrays for this purpose. The use of these data sources can make positive impact on active safety providing additional information about the vehicle and environmental condition. In this article a system for vehicle detection and determination of the arrival angle is presented. It can be used to improve the vehicle safety on road intersections where not always other vehicles can be detected by sensors such as lidars or cameras due to line of sight occlusions. The system makes use of sound source localization (SSL) methods like MUSIC, CSSM, WAVES, TOPS and SRP-PHAT which are briefly described further. The performance obtained by use of each method is compared with others. Presented results from experiments in real environmental conditions offer reasons for further work in this topic.