Abstract
Potholes are a major safety concern on roads as they often lead to accidents. Identifying them promptly is vital in preventing accidents. This research focuses on potholes that are very evident during the rainy season because These road defects pose great difficulties for drivers. This study presents the creation of an automatic pothole segmentation model for real time road damage assessment. Potholes have severe safety implications and infrastructure problems, which indicate a need for effective monitoring and maintenance strategies. A YOLOv8based segmentation model was trained using computer vision and machine learning techniques with a curated dataset of road images. Then, we fine-tuned this model through transfer learning while evaluating its performance using various metrics to detect and segment potholes accurately. After that, we integrated the model into a real time video processing pipeline which is combined with road monitoring systems so as to continuously assess the state of roads. Finally, we discuss deployment architecture, real time performance evaluation, use cases as well as future research directions towards automated pothole segmentation’s potential in enhancing road safety and infrastructure management.