Paving the Way: AI-Driven Pothole Detection with Computer Vision
Abstract
In this study, we propose a deep-learning based framework for real-time pothole detection on road surfaces. Leveraging a custom convolutional neural network trained on a comprehensive dataset of annotated road images, our method achieves over 95% accuracy in identifying pothole regions. We integrate semantic segmentation with classical image-processing filters to refine boundaries, and demonstrate deployment on an edge device for live monitoring. Experimental results show the system can process HD video at 20 FPS, making it feasible for municipal vehicle fleets to perform continuous road health assessments.