Transfer learning for LiDAR-based lane marking detection and intensity profile generation
Published in Geomatics, 2021
This paper presents a transfer learning approach using a fine-tuned U-Net model for lane marking extraction from LiDAR data, reducing the need for large training datasets. An encoder-trained model outperformed the decoder-trained version (F1-score: 86.9% vs. 82.1%), achieving 90.1% accuracy on an independent dataset.
Recommended citation: Patel, A., et al. Transfer Learning for LiDAR-Based Lane Marking Detection and Intensity Profile Generation. Geomatics 2021, 1, 287-309
Download Paper