Pothole Detection and Cost Estimation Using Deep Learning

Abstract

In this research, we present an advanced pothole detection system designed to not only identify potholes but also accurately calculate their dimensions, volume, and repair cost—all from video footage of road surfaces. The motivation behind this work stems from the need to improve road maintenance methods, which are often time-consuming and heavily dependent on manual labour. Our system leverages powerful object detection models that analyze each frame of the video to precisely identify potholes. Once detected, the system goes further by estimating the pothole’s length, breadth, and depth using a technique called monocular depth estimation. This approach cleverly allows depth to be inferred from a single camera feed, eliminating the need for expensive depth sensors or stereo vision systems. After estimating the depth, the system calculates the volume of each pothole—a crucial metric in determining the amount of material needed for repair. Based on this volume, the system automatically estimates the cost to fill each pothole, offering a full pipeline from detection to cost analysis. What sets this solution apart is its ability to automate the entire process—from detection to cost estimation—making road inspections significantly more efficient and far less labour-intensive compared to traditional manual inspections. The approach not only improves accuracy in detecting potholes but also aids in better planning and budgeting for large-scale infrastructure maintenance, paving the way for smarter cities and safer roads.