Semantic segmentation network for motorway point clouds considering reflectance and density features

Abstract Mobile LiDAR System (MLS) can quickly acquire a large amount of point cloud data of motorway scenes, and accurate semantic segmentation of the acquired MLS point clouds has a significant role in high-precision maps and digital twins. However, the MLS point cloud has unevenness in spatial distribution, the motorway scene is relatively large, and the point cloud data of street lights and other features are too sparse, resulting in a scene with relative complexity, which leads to the network sampling process to focus on the dense point cloud data. The results of semantic segmentation of the existing methods are not ideal. The MLS point cloud's semantic segmentation relies primarily on manual, repetitive labour and time-consuming work. To address the above problems, we propose a deep learning network guided by an unsupervised scene simplification module for the semantic segmentation task of motorway MLS point cloud data. We design an unsupervised module for simplifying the motorway scene to reduce the semantic segmentation burden of the deep learning model based on the MLS feature point cloud reflectivity and other characteristics of the motorway; we design a sampling method applicable to the MLS motorway scene point cloud to reduce the influence of the inhomogeneous point cloud in the sampling process and to increase the ability of the model to learn the features of the sparse features, such as the signage. The experimental results show that the overall accuracy of the method for MLS road point cloud segmentation is 0.986, and the mIoU reaches 0.937. Compared with other methods, the technique can achieve a more accurate semantic segmentation for uneven and sparse mobile LiDAR system road point clouds .