A study on source separation of traffic vehicle noise using deep learning

In order to reduce the environmental damage caused by road traffic noise, it is not only necessary to measure the amount of traffic by vehicle type and the resulting noise level but also convert them into big data for monitoring them. To achieve this, a monitoring system needs to be developed that can more accurately classify vehicle types and evaluate the noise level of each vehicle in motion. In recent years, the development of AI has significantly improved the performance of vehicle counting systems, and these systems are already being used in traffic volume surveys. However, a system that simultaneously measures traffic noise and converts it into data has yet to be developed, and the specification and prototyping of such a system is of utmost importance. This paper presents a Traffic Monitoring System, a novel system that automatically collects data on road traffic noise, which includes a noise source separation system to evaluate traffic noise, and strives to establish the related elemental technologies. Specifically, the Traffic Monitoring System uses cameras and microphones installed on edge devices to acquire road video and audio, and the system then performs vehicle detection, vehicle speed estimation, and noise evaluation for each vehicle.