AI-Based Pothole Detection for Road Safety Enhancement Using Computer Vision and Deep Learning

Abstract

Potholes can be a real headache. They not only annoy drivers but can also cause accidents, damage vehicles, and lead to costly repairs. Traditionally, spotting these pesky road hazards has meant relying on manual inspections, which can be slow and often miss the mark. But thanks to advancements in artificial intelligence and computer vision, Now we have better ways to identify potholes using rea- time video. We trained the model on a custom dataset (annotated with Roboflow) and enhanced it with data augmentation so it performs reliably in challenging lighting and weather conditions. The software performs real time analysis of live video through equipments such as webcams or dashcams, then identifies potholes in that time, and marks them with bounding boxes around and confidence scores It does records every detection made , both in a local CSV file and in a Firebase Realtime database, so the data can be centrally tracked and analyzed and well monitored. We also made sure the application is user-friendly, meaning easy to use. It provides features such as simple graphical interface built with Tkinter and OpenCV, allowing users to start the detection process, see results live, and access logged data hassle- free. With good accuracy of—92.1% precision and a mean average precision (mAP) of 93.2%—this system provides both reliable and speedy. And it’s designed to work on smaller devices like Raspberry Pi or Jetson Nano, making it compact with less space-consuming. This project can be made advanced by utilising the same in either smart cities or other put together with drones. In brief, this pothole detection system provides help to cities and huge transport authorities to automate road maintenance system, prioritizing safety over everything for everyone. By inserting the power of AI, we can sort the pothole problem more effectively and efficiently than ever before.