Automated LiDAR-Based Dataset Labelling Method for Road Image Segmentation in Autonomous Vehicles

This research introduces an automated method for labeling datasets in road segmentation, specifically tailored for autonomous vehicle technology, using LiDAR sensor data. It employs LiDAR surface contour data to pinpoint potential segmentation points. These points are further refined through a series of methods including road boundary estimation, local mapping, and a road curve fitting algorithm. We also developed a more generalized method for inverse perspective mapping using artificial neural network algorithm. Subsequently, a mask creation process is undertaken, aimed at producing datasets for camera-based systems, which are notably more cost-effective compared to LiDAR-based alternatives. The outcome of this research is a polygon-based mask dataset, specifically designed for road image segmentation. In our experiments, we successfully generated over 2000 datasets covering a 1km road segment within the ITS campus complex, achieving a mean IoU score of 0.90. This innovative approach has the potential to increase the efficiency and affordability of autonomous vehicle technology.