Detecting Hazards on Road Surfaces through Image Enhancement

This study aimed to analyze and improve the problems of the technology in order to advance the technology for detecting risks on existing road surfaces.As a major achievement, the priority of road surface risk detection and the low-light image quality improvement algorithm were developed.The image quality improvement algorithm for low-light images showed a success rate of 95%, which can improve the identification rate of images that are misrecognized due to problems such as illuminance, clarity, and brightness.In addition, the production of a prototype was completed, and the Structural Similarity Index Map verification result showed a refinement success rate of 91%.In the visual verification targeting experts, it was determined that all images had clear boundaries of identification targets after low-light image refinement.In relation to securing a learning dataset, which is one of the most important parts of the AI model, 40,000 learning datasets in low-light situations were created so that they could be used as a learning dataset.In addition, based on the results of traffic accident analysis and literature review, seven types of road condition abnormalities affecting road safety were classified, and expert hierarchical analysis was conducted to select priorities by considering three influencing factors: traffic accident occurrence,