Roadway Ice/Snow Detection using a Novel Infrared Thermography Technology

Zhu, Xuan; He, Xiangdong; Ramezanpourkami, Moein; Wu, Yuning; Zhang, Keping; Yang, Xianfeng · 2024 · ROSA P / Iowa. Department of Transportation. Aurora Program

archive: archived pipeline: cataloged verified

Get this paper ↗ (full text — opens at the source; we link to it, we don't host it)

Summary

This study addresses the critical safety challenge of detecting slippery road conditions, which contribute to 24% of weather-related vehicle crashes in the United States. Existing Road Weather Information Systems (RWIS) and remote sensors often provide only single-spot measurements that may not accurately represent broader road surface conditions. To overcome these limitations, the researchers developed a novel sensing technology capable of estimating multi-lane roadway snow coverage using dual-spectrum cameras that capture both optical and infrared imagery. This approach aims to provide reliable detection across varying illumination conditions, supporting better resource planning for snow plowing and enhancing winter traffic safety. The research was conducted over two winter seasons at field sites in Utah near US 89 and I-80. The team deployed dual-spectrum cameras (InfiRay IRS-FB462A and FLIR A50) to collect simultaneous optical and infrared images during significant snowstorms. Data preprocessing involved image registration to align the different fields of view and lane splitting to isolate individual lanes. The study evaluated two analytical approaches: traditional computer vision algorithms (including k-means clustering and Support Vector Machines) and a transfer learning framework using a U-Net architecture. The transfer learning model was specifically designed to address the constraint of limited labeled data by leveraging pre-trained knowledge, thereby reducing the need for extensive hyperparameter training. The results demonstrated that optical images were effective for snow detection during daytime conditions with sufficient illumination, while infrared images significantly outperformed optical images at night or under low-light conditions due to higher temperature contrast between snow and pavement. The transfer learning algorithm achieved a precision of 88% using daytime optical images and an impressive 94% precision using nighttime thermal images. These performance metrics were comparable to or superior to those of the traditional computer vision algorithms. The study confirmed that the dual-spectrum system could reliably distinguish snow-covered from non-snow-covered surfaces, with infrared data proving particularly robust when optical visibility was compromised. The significance of this work lies in its validation of a non-invasive, multi-lane detection system that operates effectively in diverse lighting conditions. By demonstrating high precision with limited datasets through transfer learning, the study offers a scalable solution for transportation agencies. This technology enables more accurate real-time assessments of road conditions, facilitating improved decision-making for winter maintenance operations and ultimately enhancing highway safety in regions prone to ice and snow.

Key finding

A transfer learning model using dual-spectrum imagery achieved 88% precision for daytime optical images and 94% precision for nighttime thermal images in estimating roadway snow coverage.

Methodology

field_study

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed. Discovered via bulk_ingest_rosap on 2026-05-23 (6 acquisition events logged).

Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• roadway lighting effects