Document Type

Thesis

Date of Award

10-2025

School/College

College of Science, Engineering, and Technology (COSET)

Degree Name

MS in Transportation Planning & Management

Committee Chairperson

Fengxiang Qiao

Committee Member 1

Qi Yi

Committee Member 2

Mehdi Azimi

Committee Member 3

Yachi Wanyan

Keywords

Neural network, artificial intelligence, pavement marking, multimodal datasets, convolutional detection, retroreflectors

Abstract

Pavement markings are essential roadway assets that enhance driver guidance, reduce collision risks, and support the operational reliability of autonomous and connected vehicle systems. Maintaining adequate marking visibility, particularly retro reflectivity as required by FHWA MUTCD Section 3A.05, is critical for nighttime safety and for minimizing crash severity in adverse conditions. Traditional assessment methods, including manual inspections, retro reflectometers, and service life estimates, offer useful baseline information but remain slow, labor intensive, and unsuitable for large scale or real time asset management. These limitations have increased interest in artificial intelligence solutions capable of automating pavement condition evaluation with greater consistency and operational efficiency. This study investigates the application of advanced computer vision techniques for automated pavement marking detection and classification within transportation asset management frameworks. The research objectives include evaluating the functional role of pavement markings in roadway safety and asset management; identifying limitations in current inspection practices; developing a methodological framework for data collection, preprocessing, annotation, and artificial intelligence model training; demonstrating how artificial intelligence driven detection can support predictive maintenance and cost effective asset planning; and highlighting gaps and directions for future research. The methodology uses roadway imagery obtained from NJDOT Web Straight Line Diagrams, GPS-enabled cameras, and photobooks. More than 200 images were normalized and augmented through rotations, contrast adjustments, noise injection, and flips. The dataset was split into training and test sets at an 80:20 ratio, and model performance was evaluated using precision. The study also identifies challenges, including class imbalance, environmental variability, and the need for expanded multimodal datasets to improve generalizability. Overall, the research demonstrates that artificial intelligence-driven pavement marking detection offers a strong foundation for predictive maintenance, improved safety outcomes, and long-term infrastructure resilience.

Download

Share

COinS