Dataset WAV

Nowadays, a variety of intelligent systems for autonomous driving have been developed, which have already shown a very high level of capability. One of the prerequisites for autonomous driving is an accurate and reliable representation of the environment around the vehicle. Current systems rely on cameras, RADAR, and LiDAR to capture the visual environment and to locate and track other traffic participants. Human drivers, in addition to vision, have hearing and use a lot of auditory information to understand the environment in addition to visual cues. In this research, we present the sound signal processing system for auditory based environment representation.

Sound propagation is less dependent on occlusion than all other types of sensors and in some situations is less sensitive to different types of weather conditions such as snow, ice, fog or rain. Various audio processing algorithms provide the detection and classification of different audio signals specific to certain types of vehicles, as well as localization.

The ambient sound is classified into fourteen major categories consisting of traffic objects and actions performed. Additionally, the classification of three specific types of emergency vehicles sirens is provided.

The presented approaches and methods have great potential to increase the accuracy of environment perception and, consequently, to improve the reliability and safety of autonomous driving systems in general.