A Novel Framework for Road Information Extraction From Low-Cost MMS Point Clouds

Abstract

In this study, a novel framework was developed and presented for the extraction of comprehensive road information from low-cost mobile mapping system data, addressing the needs of various applications. The methodology begins with an iterative weighting method to accurately identify ground points, followed by a machine-learning-integrated approach for road boundary detection and road surface segmentation. The extracted road boundary points were then used to calculate key geometric parameters, including cross-slope, longitudinal slope, elevation change, and road width. Finally, road markings were extracted using the RGB features of point cloud data from the MMS system. The results showed that the mean absolute error for longitudinal slope in the forward and return directions was 0.1% and 0.08%, respectively, while the cross-slope values exhibited deviations of 0.19% and 0.21% compared to the reference data. Road markings were extracted using the RGB features of MMS data, achieving a recall of 96.42%, precision of 94.75%, and an F1-score of 95.58%. Comparative analysis revealed that the proposed approach outperformed conventional image-based methods, with an average deviation of 2.9 cm from reference data in lane line detection. Overall, this workflow successfully identifies critical information such as precise road boundaries, road markings, and road geometry, demonstrating the potential of MMS data as a reliable and cost-effective alternative for detailed road information analysis.