With the rapid advancement of smart railway construction, semantic segmentation technology based on three-dimensional (3D) point cloud data has become a core enabler for the intelligent operation and maintenance of railway facilities. Railway scenes present significant challenges for point cloud segmentation, including highly irregular and unstructured data, severe occlusions leading to information loss, and extreme scale imbalance among targets ranging from large catenary pillars to small fasteners. This paper provides a comprehensive review of 3D point cloud semantic segmentation methods tailored for complex railway environments. It systematically analyzes the unique characteristics of railway point cloud data and the resulting algorithmic challenges. The review covers both traditional and deep learning–based segmentation approaches, evaluating the performance and adaptability of different neural network architectures in meeting railway-specific requirements such as track continuity maintenance, small component recognition, and noise robustness. Key benchmark datasets and evaluation metrics are also discussed. The analysis highlights that future research should prioritize noise handling, feature enhancement, multimodal data fusion, and the development of lightweight model architectures to achieve robust real-time segmentation in complex and dynamic railway settings.
石 et al. (Mon,) studied this question.