Abstract:Point cloud ground segmentation is a key preprocessing task in Mobile Laser Scanning (MLS)-based measurement and sensing. However, ground segmentation currently faces major challenges such as diverse ground morphology, sparse point cloud data, and interference from reflection noise. Meanwhile, since the existing Principal Component Analysis (PCA)-based ground plane fitting methods lack the judgment of iterative convergence and automatic correction of non-ground plane fitting results, this not only leads to unnecessary computational overhead, but also affects the accuracy of ground segmentation. To address these issues, this paper proposes a three-stage MLS point cloud ground segmentation method based on ground plane fitting, called GPF-Plus. This method adopts a three-stage strategy based on ground plane fitting to achieve ground segmentation, which is able to effectively deal with the challenges of various terrains. Firstly, the Initial Ground Segmentation of the Original Point Cloud is performed to quickly produce a coarse segmentation result. Secondly, the False Negative Points Extraction is performed to improve the recall. Finally, the False Positive Points Extraction is performed to improve the precision. At the same time, the Infinite Polar Grid Model is used to divide the point cloud, which reduces the number of grids and effectively alleviates the problem caused by point cloud sparsity. The Reflection Noise Removal Mechanism is introduced to enhance the robustness to reflection noise. In addition, the Improved Ground Plane Fitting improves the accuracy and speed of ground plane fitting. In this paper, experimental validation is carried out using the SemanticKITTI dataset, the SimKITTI32 dataset, and the collected point clouds of the mine environment. Compared with the state-of-the-art methods, GPF-Plus has excellent accuracy, real-time performance and robustness, and has high application potential in the field of measurement and sensing.

Multispectral point cloud superpoint segmentation

Superpoint-guided Semi-supervised Semantic Segmentation of 3D Point Clouds

Multi-Scale Superpoint Network for 3D Point Cloud Semantic Segmentation.

PointMS: Semantic Segmentation for Point Cloud Based on Multi-scale Directional Convolution

Superpoint Network for Point Cloud Oversegmentation

Efficient LiDAR Point Cloud Oversegmentation Network

SSPC-Net: Semi-supervised Semantic 3D Point Cloud Segmentation Network

Learning Inter-Superpoint Affinity for Weakly Supervised 3D Instance Segmentation

Associate Semantic-Instance Segmentation of 3D Point Clouds Based on Local Feature Extraction

PointGT: A Method for Point-Cloud Classification and Segmentation Based on Local Geometric Transformation

FreePoint: Unsupervised Point Cloud Instance Segmentation

PointGS: Bridging and Fusing Geometric and Semantic Space for 3D Point Cloud Analysis

Semasuperpixel - A Multi-Channel Probability-Driven Superpixel Segmentation Method.

An efficient point cloud semantic segmentation network with multiscale super-patch transformer

GPF-Plus: A Three-Stage MLS Point Cloud Ground Segmentation Method Based on Ground Plane Fitting

Fast Multiscale Superpixel Segmentation for SAR Imagery

PointMM: Point Cloud Semantic Segmentation CNN under Multi-Spatial Feature Encoding and Multi-Head Attention Pooling

A Normalized Spatial–Spectral Supervoxel Segmentation Method for Multispectral Point Cloud Data

Learning Superpoint Graph Cut for 3D Instance Segmentation

Deep Learning on Superpoint Generation with Iterative Clustering Network

Point Cloud Segmentation based on Hypergraph Spectral Clustering