Enhanced Road Safety: Real-Time Pothole Detection Using YOLOv9

Abstract

The goal of this research is to develop a real time pothole detection system with the advanced  YOLOv9 model which outperforms previous ones, such as YOLOv5  and YOLOv8. PGI and GELAN in YOLOv9 enhance  the accuracy of the detection through programmable gradient information (PGI) and the Generalized Efficient Layer  Aggregation Network (GELAN), which reduce data loss and improve the performance. The methodology involves  training the YOLOv9 model on a large dataset of diverse images and video streams which cover  a wide range of road conditions and lighting. The results show that the YOLOv9 model  has a detection accuracy over 90% with real time processing speed of up to 31.76  frames per second (FPS). This study concludes that such advanced technology can be very useful in improving road  safety by enabling faster repair and more effective maintenance, which in the end can help solve some of the  major challenges in infrastruture management.