Home / Current Issue / Paper 1717175
Yolo V8 And Canny: A Dual Detection System for Road Defect Analysis and Lane Guidance
Subject area: Science,Engineering and Technology · Area of research: Computer Vision & Road Safety
DOI: https://doi.org/10.64388/IREV9I10-1717175
Abstract
Road safety systems need to do two things at once: spot hazards on the surface and track where the lane boundaries are. Most approaches handle one or the other. We built a system that does both. The framework pairs a YOLOv8m deep learning model for road damage detection with a classical computer vision pipeline for real-time lane tracking. Training followed a global-to-local strategy — we first trained on 6,832 images from India and Japan to build a broad foundation, then fine-tuned specifically on US road data. This transfer learning approach lets the model handle regional variation without starting from scratch for each country. For lane detection, we used Canny edge detection and Probabilistic Hough Transforms, constrained by a dynamic spatial mask that focuses the system on the relevant driving area. The result: a fine-tuned model with an mAP@0.5 of 0.571 across multiple road distress categories. A FastAPI backend ties both pipelines together, enabling concurrent damage detection and lane guidance in real time. Combining deep learning with geometric constraints gives autonomous systems a more complete picture of the road than either method could alone.
How to cite this paper
@article{1717175,
author = {Madala jayanth Kumar, Patti Ebbeju, Maddineni Sai Ganesh, Kesana Pawan Chandhu},
title = {Yolo V8 And Canny: A Dual Detection System for Road Defect Analysis and Lane Guidance},
journal = {Iconic Research And Engineering Journals},
year = {2026},
volume = {9},
number = {10},
pages = {3297-3301},
issn = {2456-8880},
url = {https://www.irejournals.com/formatedpaper/1717175.pdf},
abstract = {Road safety systems need to do two things at once: spot hazards on the surface and track where the lane boundaries are. Most approaches handle one or the other. We built a system that does both. The framework pairs a YOLOv8m deep learning model for road damage detection with a classical computer vision pipeline for real-time lane tracking. Training followed a global-to-local strategy — we first trained on 6,832 images from India and Japan to build a broad foundation, then fine-tuned specifically on US road data. This transfer learning approach lets the model handle regional variation without starting from scratch for each country. For lane detection, we used Canny edge detection and Probabilistic Hough Transforms, constrained by a dynamic spatial mask that focuses the system on the relevant driving area. The result: a fine-tuned model with an mAP@0.5 of 0.571 across multiple road distress categories. A FastAPI backend ties both pipelines together, enabling concurrent damage detection and lane guidance in real time. Combining deep learning with geometric constraints gives autonomous systems a more complete picture of the road than either method could alone.},
month = {April},
doi = {https://doi.org/10.64388/IREV9I10-1717175}
}