Comparative Analysis of Deep Learning Architectures for Automated Pothole Detection and Segmentation in Road Infrastructure Monitoring

View at publisher Summarise with VIRAT
Authors: Shravan A P, M. Phani Nithin, Sistla Sai Sankalp, Gopika Madambath, Karthika. R
Cited by: 0
References: 19

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

In developing countries like India, the role of road infrastructure in enabling daily mobility and facilitating logistical management through the logistics supply chain is significant. However, road maintenance is predominantly conducted manually, is a time-consuming process, and requires an extensive amount of effort from field workers. Continual water entry and frequent vehicle loads on roads will create potholes. These potholes create hazards for both pedestrians and automobiles and disrupt the flow of traffic through the region. With the increased use of autonomous vehicles and advanced driving assistance systems, accurately identifying and locating potholes has become critical. This paper presents a comparison of five deep learning models for identifying and segmenting potholes, specifically YOLOv11, YOLOv8n, the Vision Transformer (ViT), the Data-Efficient Image Transformer (DeiT), and Segment Anything Model (SAM), under different environmental conditions (urbanrural). A publicly available dataset containing the ground truth annotation information was used to conduct the evaluation of the five deep learning models on their suitability for use in real-life situations. The evaluation of each model is based upon three deployment parameters: (1) accuracy; (2) inference speed; (3) computational efficiency. The results of this study are intended to provide practical insights into the appropriateness of these models for real-world applications, as well as provide information regarding the trade-offs between model accuracy and resource constraints related to the automated monitoring of road conditions.

Fields of Study & Subjects

Infrastructure Maintenance and Monitoring Advanced Neural Network Applications Industrial Vision Systems and Defect Detection

Publication details

Journal / source
International Conference on Intelligent Systems for Communication, IoT and Security (ICISCoIS)
Pages
1–8
Publication date
8 January 2026
Language
EN