A novel algorithm for key road sections obstacle detection based on LiDAR

Abstract

Ensuring the safety of key road sections is crucial, and high-precision Obstacle Detection (OD) is necessary to achieve this goal. In this study, we utilized LiDAR technology to develop a novel algorithm that accurately detects obstacles in critical road sections. The algorithm consists of three main parts: Firstly, the point cloud is segmented into scan lines, and the distribution error of the scan lines is used to filter out point clouds that correspond to road boundary. Secondly, we used an improved DBSCAN algorithm to achieve global robust clustering. This method introduces a modulation function to dynamically regulate the neighbourhood radius during the clustering process by referring to the density distribution characteristics of point clouds in the $x$ and $y$ directions. This step was employed to extract suspicious obstacle clusters. Finally, we extract regions corresponding to the real-time input point cloud from the reference point cloud. Segmented ICP (Seg-ICP) registration was performed in the global scope to obtain a true set of obstacle clusters. Experiment results in multiple locations demonstrated that our algorithm could effectively identify dangerous obstacles in key road sections. Our algorithm can detect obstacles of 10 cm * 10 cm * 10 cm within a range of 50m and its Average Final Detection Rate (AFDR) can reach at 99.47%, providing an optional means of safety protection for key road sections.