Mobile Mapping Mesh Change Detection and Update

Mobile mapping, in particular, Mobile Lidar Scanning (MLS) is increasingly widespread to monitor and map urban scenes at city scale with unprecedented resolution and accuracy. The resulting point cloud sampling of the scene geometry can be meshed in order to create a continuous representation for different applications: visualization, simulation, navigation, etc. Because of the highly dynamic nature of these urban scenes, long term mapping should rely on frequent map updates. A trivial solution is to simply replace old data with newer data each time a new acquisition is made. However it has two drawbacks: 1) the old data may be of higher quality (resolution, precision) than the new and 2) the coverage of the scene might be different in various acquisitions, including varying occlusions. In this paper, we propose a fully automatic pipeline to address these two issues by formulating the problem of merging meshes with different quality, coverage and acquisition time. Our method is based on a combined distance and visibility based change detection, a time series analysis to assess the sustainability of changes, a mesh mosaicking based on a global boolean optimization and finally a stitching of the resulting mesh pieces boundaries with triangle strips. Finally, our method is demonstrated on Robotcar and Stereopolis datasets.