Urban-Scale Semantic Segmentation Using PointMamba and Mobile Laser Scanning Point Clouds

Abstract. Point cloud semantic segmentation is a critical task in autonomous driving and digital twin applications. This study introduces a novel semantic segmentation approach leveraging the PointMamba network, specifically designed to address the challenges of complex urban scene point cloud data. The PointMamba network integrates a state space model (SSM) with point cloud serialization and advanced feature extraction techniques, yielding significant performance improvements in semantic segmentation tasks. PointMamba was rigorously evaluated on the Toronto3D urban scene point cloud dataset, achieving an Overall Accuracy (OA) of 93.94% and a mean Intersection over Union (mIoU) of 66.03%. Comparative studies demonstrated that PointMamba outperformed existing point-based methods, including PointNet++ and PointNet, in handling intricate urban environments, delivering superior semantic segmentation outcomes on complex urban road environments.