2025 Volume 11 Issue 2 Pages A_71-A_79
While the number of traffic accidents has been decreasing yearly, the reduction rate for accidents on residential roads is smaller compared to arterial roads. Predicting high-risk intersections on residential roads is thus a crucial challenge for reducing traffic accidents in these areas. In this study, we aim to explore the applicability of a model based on Convolutional Neural Networks (CNN), a rapidly advancing field of image recognition AI, to predict intersections with a high likelihood of traffic accidents. Additionally, we aim to identify the optimal conditions for intersection images used in the prediction model. The implementation and evaluation of the model reveal that image recognition AI is indeed applicable to predicting traffic accidents. The optimal image conditions are determined to be a resolution of 1,280×960 pixels, a sky color of uneditted, and a distance of 10 meters from the intersection. Furthermore, it is found that the presence of highly saturated buildings in the images significantly influences the recognition accuracy of the AI.
