A diffusion model-based intelligent optimization method of rural road environments

Abstract

Well-designed rural road environments can guide drivers to adopt reasonable driving behaviors, thereby significantly improving the driving experience and ensuring road safety. Existing methods for optimizing rural road environments mainly rely on expert knowledge, have low automation degrees, and are limited in efficiency and accuracy. Therefore, this study aims to propose an intelligent optimization method for rural road environments by using image generation technology. Using environment images from a naturalistic driving dataset, the area and location information of semantic components (e.g., lane markings, vegetation, guardrails, and traffic signs) in rural road environments are extracted, and their impacts on driving speed are analyzed based on explainable machine learning (extreme gradient boosting (XGBoost) and Shapley additive explanations (SHAP)). These impacts are then utilized to determine how to adjust and optimize the road environment components at appropriate locations (i.e., obtain the optimization scheme). Then, a novel image generation technique, Diffusion model, is employed to establish an intelligent optimization method, which can directly generate optimized images of rural road environments. Compared with traditional manual mapping and other popular image generation algorithms such as CycleGAN, the method proposed in this study has the advantages of high efficiency, labor saving, and superior image generation quality. This study can facilitate the design and optimization of rural road environments and enhance rural road safety in a more intelligent way.