Towards Robust LIDAR Lane Clustering for Autonomous Vehicle Perception in ROS 2

From LIDAR pointclouds traffic lanes, racetracks, parking lanes can be extracted with clustering algorithms. However, standard clustering algorithms like DBSCAN, K-means, and BIRCH may exhibit limited robustness in recognizing these specific geometric patterns. The current paper proposes a modification of the well-known DBSCAN algorithm which is designed for autonomous vehicle lane detection. The main idea of the proposed work is to add extra steps into the classic DBSCAN algorithm, thus regulate the cluster expansion. This modification introduces some challenges too, their subsequent resolution will be addressed in detail. To reproduce our work, both the dataset and the accompanying source code in python is shared publicly.