Potholes pose significant safety hazards for drivers and can rack up hefty repair bills for cities. Enter X-PotholeNet Pro, a solution crafted to address this problem by utilizing two finely-tuned YOLOv8 models alongside cutting-edge machine vision techniques. This system not only spots potholes but also assesses their severity, categorizing them as Low, Medium, or High through a combination of three specialized classifiers. It takes into account various features like size, shape, shadow patterns, road position, surface contrast, and the surrounding context to ensure a precise evaluation. Before diving into processing, a quality check sifts through the inputs to weed out any poor-quality images, such as those that are dark, blurry, or irrelevant. Each identified pothole is marked with visual annotations and comes with straightforward explanations of its risk score, making the system both transparent and user-friendly. The platform accommodates various input types, including images, live webcam feeds, recorded videos, and uploaded footage, all easily accessible via a Streamlit dashboard. Results are generated in a flash and can be exported in JSON format for further analysis. Once installed, the system operates offline, eliminating the need for cloud services. In real-world tests conducted on the roads of Bengaluru, six potholes were detected, all receiving the highest risk score. This makes the system invaluable for road maintenance, enhancing safety for autonomous driving, and assessing vehicle damage.

Keywords—pothole detection; YOLOv8; severity classification; road safety; ensemble learning; explainable AI; computer vision; real-time detection; deep learning; CLAHE.