On the Task of Localization of Road Infrastructure Objects Based on Audio Signals

More and more effort is being invested in the development of unmanned vehicles. Often video cameras, lidars, radars, etc. are used. However, there are situations on the road when an object is not visible on the video, although it can be heard. Therefore, this article is devoted to the development of a system to build a map of objects based on audio signals, as well as determining the distance to objects. At the beginning of the article a review of existing methods of detection and localization of objects by acoustic methods is given. A bench installation for collecting a dataset of sounds on public roads is described. This paper discusses a method of object detection based on audio signals of road infrastructure. The architecture and software implementation of the audio signal collection and processing platform is given. Load testing is given at the end, as well as a description of the full-scale experiment and the results are given.