Road traffic noise map generation using aerial photographs and machine learning algorithm

Road traffic noise is a major source of environmental noise and poses significant public health risks. In many countries, predictive noise mapping is used to assess such noise exposure, typically by calculating and visualizing noise distribution using prediction models. In Japan, however, road traffic noise maps are typically created using field surveys, which are both costly and labor-intensive. To address these challenges, this study proposes a methodological approach that focuses on determining sound power levels, a fundamental step in creating accurate predictive road traffic noise maps. Traffic distribution data—including vehicle types and positions—were automatically extracted from aerial photographs using an object detection algorithm and used to estimate sound power levels based on the ASJ RTN-Model 2018. We evaluated the detection accuracy of the model and analyzed its effect on the estimation. The estimation results were further validated by comparing calculated sound pressure levels along the road and within a building complex with on-site measurements. Furthermore, using the calculated sound power levels and traffic data, a road traffic noise map for a district in Tokyo was generated.