Article ID: e26.15
Building upon our previous study on the generation of road traffic noise maps using vehicle information extracted from aerial photographs, this study focuses on the influence of vehicle recognition performance on the estimation of road traffic noise. While the earlier work demonstrated the feasibility of estimating road traffic noise distributions from aerial photographs, the effects of recognition accuracy on estimation performance were not yet examined in detail. In practice, recognition errors primarily lead to missed detections, which distort inferred traffic conditions by reducing vehicle density and inflating average running speed. To quantify the resulting estimation deviation, a simulation-based framework was developed by defining ideal recognition scenarios and generating degraded recognition scenarios with controlled recognition rates across multiple baseline average running speed conditions. The impacts of recognition errors were analyzed both in terms of overall recognition rate and vehicle-class-dependent recognition performance for light and heavy vehicles. The proposed framework was further examined using a real urban road in Tokyo by comparing manually labeled aerial-photograph data, treated as the ideal recognition scenario, with actual degraded recognition scenarios obtained by applying a YOLO model to the original aerial photographs.