Noise Model-Based Line Segmentation for Plane Extraction in Sparse 3-D LiDAR Data

Planar features serve as an important component in point cloud registration and reconstruction. However, extracting planes in the point clouds collected by a 3-D LiDAR sensor is still a challenging task due to the sparse property. To obtain reliable plane segmentation results, it is very necessary to fully exploit the scanning pattern of the sensor. In this article, we propose a novel plane extraction method for 3-D LiDAR data in a framework of point-to-line-to-plane. In the point-to-line stage, a new flat-point detector is introduced to obtain line segments. In the line-to-plane stage, we present the line based Douglas-Peucker (LBDP) algorithm to find coplanar line segments. Unlike region growing, which is generally applied to grouping line segments, LBDP does not suffer from the poor geometry of the selected initial region. More importantly, since the collected point clouds are always noisy, we model the measurement noise via statistical analysis, and bridge the noise level and parameter uncertainty to provide reasonable thresholds throughout our method. In the experiments, we evaluate the proposed method on both simulated and real datasets in terms of true positive rate (TPR), positive predictive value (PPV),
$F1$
score, and five segmentation metrics. The results show that the proposed method can accurately extract planes in real time and outperforms the compared approaches.