About Choosing the Audio Signal Representation in the Task of Vehicle Pattern Recognition

Currently, the main focus in the development of self-driving vehicles is the development of visual pattern recognition systems. However, in real operation there are situations that cannot be identified by video images, such as approaching motorcycles, cars with special signals, sound alerts, etc. To solve this problem it is necessary to develop an algorithm that allows to distinguish audio patterns. This paper discusses methods for representing audio signals and traffic conditions. It describes the dataset collection layout, its structure, and an example of the collection of audio traffic information. The method of audio signal representation is described, and a robust method of sound pattern recognition based on machine learning algorithms is proposed. Conclusions about the applicability of the developed method are made, and further tasks for the development of this topic are considered.