Artificial Intelligence and Mobile Phone-Based Pavement Marking Condition Assessment and Litter Identification

Kuang, Biao; Chen, Jianli · 2025 · ROSA P / Center for Transformative Infrastructure Preservation and Sustainability (CTIPS) Region 8 University Transportation Center (UTC)

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 inefficiencies of manual transportation asset inspections by developing an automated, artificial intelligence (AI) system for assessing pavement marking conditions and identifying roadside litter. Traditional inspection methods are labor-intensive, subjective, and costly, while previous AI efforts were limited by small datasets (~1,000 images) and lacked capabilities for object counting and precise geolocation. The research aims to create a scalable, end-to-end solution that detects, counts, and geolocates infrastructure issues to support proactive maintenance planning and improve roadway safety. The methodology involved collecting over 6,000 self-collected images for each asset type using mobile phones mounted on vehicle windshields, capturing forward-facing video and GPS data. The dataset included faded white and yellow pavement markings, as well as four classes of roadside litter: white litter, black litter, leaves, and dirt. The researchers utilized the You Only Look Once (YOLO) deep learning architecture to train two detection models. To address the challenge of duplicate detections in video sequences, an object tracking algorithm was developed to count distinct objects within specific road sections. Furthermore, a geolocation model was created to synchronize heterogeneous GPS sampling rates with video frames using time-based interpolation, enabling precise spatial mapping of identified objects. The results demonstrate strong performance across all developed modules. The pavement marking detection model achieved an F1-score of 0.88, while the roadside litter detection model achieved an F1-score of 0.84. The object counting algorithm successfully quantified distinct issues within highway segments, validated through a case study in Utah. The geolocation model exhibited high spatial accuracy, with an average positional error of only 0.27 meters when validated against ground truth data on Interstate 15. Additionally, an interactive mapping interface was implemented using the Folium Python package to visualize object class, inspection time, geolocation, and cropped images, providing an intuitive tool for asset management. The significance of this work lies in its expansion of AI capabilities from simple detection to comprehensive inspection workflows including counting and high-precision geolocation. By leveraging accessible mobile phone technology and robust deep learning models, the study provides transportation agencies with a cost-effective, scalable tool for frequent infrastructure monitoring. This approach reduces reliance on manual labor, enhances the accuracy of maintenance planning, and ultimately contributes to improved roadway safety by ensuring timely identification and remediation of faded markings and hazardous litter.

Key finding

The developed AI models achieved F1-scores of 0.88 for pavement marking issues and 0.84 for roadside litter, while the integrated geolocation model demonstrated high spatial accuracy with an average positional error of 0.27 meters.

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.

• signage environment