Pothole Detection and Instance Segmentation Using Yolo V8

Potholes on roadways are a major safety problem and can cause accidents and car damage if not corrected immediately. This research suggests an instance segmentation approach using the ability of YOLOv8 to detect potholes with deep learning. A dataset of 3,074 images of roads was obtained and marked with the instances of pothole. The dataset was used in the training of the model, that is, the YOLOv8m architecture belonging to the Ultralytics package that was optimized for the real-time detection and segmentation task. Computer vision is actually the field that provides one the ability to automate operations and manipulate multidimensional data. It is a model that, when integrated with artificial intelligence, replicates the human visual system. The model had a box mAP50 of 82.1 %, mask mAP50 of 80.5%, precision at 82.2%, and recall at 74.4%. This reflects its capability to identify potholes and segment well with great accuracy under different road conditions. This research makes significant contributions toward the automation of road maintenance by providing a scalable, efficient, and accurate method for pothole detection, which can eventually support the initiation and aid of the autonomous vehicles and road management systems. Further improvements are discussed with respect to environmental robustness and real-time deployment.