1373533 - Improving Reliability of Roadway Asset Data Collected by Vehicle-Based Platforms

Roadside asset data such as line striping, traffic signs, pavement messages, sidewalks, medians, manholes, etc. is of high importance for public transportation agencies to maintain a healthy and efficient transportation network. In recent years, collection, inventorying and assessing these assets have become of growing importance to improve the overall transportation infrastructure alongside pavements. Recording and monitoring assets has been traditionally conducted through manual field inspections which requires considerable time and cost. In past decade, there has been increasing interest in automated asset data collection methods using LiDAR (Light Detection and Ranging) and extraction of the assets using machine learning and artificial intelligence techniques. Given the volume of data generated and inherent quality issues from automated extraction methods, an efficient and comprehensive quality control process is required to generate reliable data. This paper presents a statistical-based approach for conducting comprehensive quality control on transportation asset data and improving reliability of the extracted data. In this approach, libraries of data patterns were created and programmed for each specific asset. The library of patterns continues to develop as more data is available. The extracted raw data is analyzed and compared to these rules and patterns. The sections that do not fit the guidelines of the set patterns are flagged for further review. This approach is a systematic approach to review the entire dataset and flag candidates for manual review rather than selecting random sections for quality control purposes. This study in based on an implementation which includes 25 roadway asset classes for approximately 3000 lane miles of roadway.