Automated large scale indoor reconstruction using vehicle survey data

摘要

Indoor reconstruction is important in location based service with the development of web Geographic Information System (GIS). 3D visualization virtual navigation in indoor scene is high demanded in the internet. Due to lack of texture, active range is used for location and mapping, such as structure of light and time of flight. Laser scanner is more accurate for large scale indoor mapping. A backpack or trolley mounted laser scanners and cameras are widely used in indoor mapping. Point cloud of laser scanner is mainly used to location. Camera images are used to make panoramic images. In a large scale environment, the point cloud and images are a large dataset to be shown. Indoor reconstruction can reduce the dataset in the web. This paper presents an automatic pipeline of indoor reconstruction for the dataset captured by Navvis M3 trolley. Firstly, Poisson reconstruction is used to generate the triangle mesh from the point cloud. Secondly, the mesh is partitioned into several blocks using the building principal direction to improve the efficiency, and then the mesh is simplified using plane segmentation based method. At last, a ray intersection based method is used to select the corresponding images to the mesh, and then texture atlas is generated. Experiment shows that, the pipeline can automatic get the indoor reconstruction mesh with texture. The result reduces the data size sharply, and is also convenient to be shown on the web.