Abstract: For the task of tunnel surface monitoring, point clouds obtained by 3D laser scanning suffer from undesirable characteristics such as enormous data volumes, unstructured organization, and narrow linear non-uniform distribution, which imposes significant pressure on tunnel point cloud data processing and constrains the development of tunnel monitoring application. This paper presents a tunnel massive point cloud management method based on a dual-layer indexing structure. A Hough transform-based method for preliminary determination of the horizontal tunnel centerline is designed to guide the automatic segmentation of tunnel point clouds along the centerline. Then, a “bottom-up” merging strategy is suggested for generating local octrees for the segmented point clouds; Based on this, application of the non-redundant Level of Detail (LOD) modeling and the dynamic memory dynamic scheduling enables real-time visualization of massive point clouds. The experimental results show that our new method improves efficiency significantly in extracting the tunnel's horizontal axis, and it is better than traditional approaches in point cloud retrieval and visualization of massive point cloud data.

Key words: massive point cloud, tunnel engineering, big data processing, dual-layer indexing structure, internal and external memory dynamic scheduling