Urban Road Anomaly Monitoring Using Vision–Language Models for Enhanced Safety Management DOI Creative Commons

Hanyu Ding,

Yawei Du, Zhengyu Xia

et al.

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(5), P. 2517 - 2517

Published: Feb. 26, 2025

Abnormal phenomena on urban roads, including uneven surfaces, garbage, traffic congestion, floods, fallen trees, fires, and accidents, present significant risks to public safety infrastructure, necessitating real-time monitoring early warning systems. This study develops Urban Road Anomaly Visual Large Language Models (URA-VLMs), a generative AI-based framework designed for the of diverse road anomalies. The InternVL was selected as foundational model due its adaptability this purpose. URA-VLMs features dedicated modules anomaly detection, flood depth estimation, level assessment, utilizing multi-step prompting retrieval-augmented generation (RAG) precise adaptive analysis. A comprehensive dataset 3034 annotated images depicting various scenarios developed evaluate models. Experimental results demonstrate system’s effectiveness, achieving an overall detection accuracy 93.20%, outperforming state-of-the-art models such InternVL2.5 ResNet34. By facilitating decision-making, AI approach offers scalable robust solution that contributes smarter, safer environment.

Language: Английский

Urban Road Anomaly Monitoring Using Vision–Language Models for Enhanced Safety Management DOI Creative Commons

Hanyu Ding,

Yawei Du, Zhengyu Xia

et al.

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(5), P. 2517 - 2517

Published: Feb. 26, 2025

Abnormal phenomena on urban roads, including uneven surfaces, garbage, traffic congestion, floods, fallen trees, fires, and accidents, present significant risks to public safety infrastructure, necessitating real-time monitoring early warning systems. This study develops Urban Road Anomaly Visual Large Language Models (URA-VLMs), a generative AI-based framework designed for the of diverse road anomalies. The InternVL was selected as foundational model due its adaptability this purpose. URA-VLMs features dedicated modules anomaly detection, flood depth estimation, level assessment, utilizing multi-step prompting retrieval-augmented generation (RAG) precise adaptive analysis. A comprehensive dataset 3034 annotated images depicting various scenarios developed evaluate models. Experimental results demonstrate system’s effectiveness, achieving an overall detection accuracy 93.20%, outperforming state-of-the-art models such InternVL2.5 ResNet34. By facilitating decision-making, AI approach offers scalable robust solution that contributes smarter, safer environment.

Language: Английский

Citations

0