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

Abstract

Pavement markings are an important traffic control device, enhancing both the safety and efficiency of various modes of transportation by aiding vehicles, bicyclists, and pedestrians in effectively navigating the transportation network. In this PacTrans project, we developed a new framework to extract and classify various road markings from mobile laser scanning (MLS) data. The proposed framework consists of three principal steps: road surface extraction, road marking extraction, and road marking classification. For road surface extraction, using geometric information from the point cloud, ground filtering followed by slope filtering are applied to extract a road surface that is likely to include road markings. Next, the extracted road surface point cloud data are rasterized with Otsu’s method into 2D to generate an intensity image and segment high-intensity pixels, likely representing road markings. To reduce false positives while preserving the actual road markings, we apply the block partition and high-pass filtering approaches. Finally, for road marking classification, common linear lane markings with lengths greater than a predefined threshold are first segmented, and then remaining markings are fed into a template matching program for classification. The developed program was evaluated by using a variety of MLS data collected by the Oregon Department of Transportation (ODOT). The experimental results showed that the developed program outperformed our previous version of the road marking extraction tool by extracting highly curved and complex road markings with significantly fewer false positives. The developed program can be used to support informed decision making by state transportation agencies for effective management of road markings.