Intensity thresholding and deep learning based lane marking extraction and lane width estimation from mobile light detection and ranging (LiDAR) point clouds

This study enhances lane marking extraction from LiDAR point clouds using unsupervised intensity normalization and deep learning with automated labeling. Compared to traditional intensity thresholding, normalization improves extraction accuracy (F1-score: 78.9% vs. 72.3%), while deep learning trained on auto-labeled data outperforms manually labeled models (85.9% vs. 75.1%). Both deep learning approaches achieve better lane width estimation and gap detection, enabling more effective automated road monitoring and maintenance.

Recommended citation: Patel, A., et al. (2020). Intensity Thresholding and Deep Learning Based Lane Marking Extraction and Lane Width Estimation from Mobile Light Detection and Ranging (LiDAR) Point Clouds. Remote Sensing, 12(9), 1379.
Download Paper