Guardrail Detection in Rural Roads Using Zero- and Few-Shot Learning with Vision Language Models

Document Type

Conference Proceeding

Publication Date

1-1-2026

Abstract

The high incidence of rural crashes highlights the critical need to analyze roadside safety characteristics on rural roads. Among these characteristics, the presence of guardrails is recognized as a key safety countermeasure. Existing vision-based methods for detecting guardrails primarily rely on supervised object detection using deep convolutional neural networks (CNNs), which require large amounts of annotated data and have been criticized for their limited generalizability. To address this challenge, this paper proposes a vision-based approach for automatically detecting guardrails in rural road images by leveraging zero-shot and few-shot learning techniques with various Generative Pre-trained Transformer (GPT) models using both reasoning and chains of thought, including state-of-the-art models such as GPT-4o, GPT-4-mini, and GPT-4-1. The results demonstrate that the proposed approach outperforms existing methods in terms of accuracy, achieving a 99% detection rate for guardrails. Compared to manual inspection and vision-based methods using CNNs, the proposed method offers notable advantages, including performance and ease of implementation.

Publication Title

Construction Research Congress 2026 Advanced Technologies Artificial Intelligence and Data Analytics in Construction Selected Papers from Construction Research Congress 2026

ISBN

[9780784486962]

Link to Full Text

Share

COinS