pacific nw transportation consortium region 10 Pactrans

Efficient Extraction and Evaluation of Complex Pavement Markings from Mobile Laser Scan Data

PI: Michael Olsen (OSU), olsen@oregonstate.edu, ORCID: 0000-0002-2989-5309

Co PIs: none

AMOUNT & MATCH: $30,000 from PacTrans; $30,000 Match

PERFORMANCE PERIOD: 8/16/2018 – 8/15/2020

STATUS: Completed

CATEGORIES: Pavement Markings, Laser Scanning

RESEARCH PROJECT HOT SHEET:

UTC PROJECT DOCUMENTATION:

FINAL PROJECT REPORT:

PROJECT DATA:

DESCRIPTION: Recently, in our research with the Oregon Department of Transportation (ODOT), the research team developed an automated method for extracting linear lane markings from MLS data (Figure-a) as well as evaluating the retroreflectivity of those markings. In the current Pactrans project we are building upon that effort to develop advanced techniques to handle more complex markings (e.g., pedestrian crosswalks, chevrons, and arrows) that were not considered in the prior project, but important to support mobility for multi-modal transportation. First, we project the MLS data into 2D to generate an intensity image and segment high intensity pixels, likely representing various road markings (Figure-b). Subsequently, a deep learning neural network approach, which is known for its high performance for object recognition in many applications, is used to classify various types of markings. This research will enable performance-based procedures for transportation agencies to evaluate pavement marking quality by providing detailed information, including retroreflectivity and types of markings, ranging from high resolution data on a single stripe to aggregated data and analyses statewide. This, in turn, supports informed decision making by DOT management for effective resource allocation. Improved maintenance of pavement markings will also lead to improved mobility with technologies such as autonomous vehicles.