Automatic object segmentation and reconstruction in LIDAR point clouds of railway environments

Abstract

Point clouds are very valuable in GIS and can be used to extract many kinds of information from an environment. However, there are two main shortcomings of unprocessed point clouds: they are not very efficient in visualizations, and it is hard to visually discern between objects. This thesis presents an automatic method for segmenting and reconstructing objects inside point clouds in the context of railway environments. To achieve a more efficient visualization and better discernibility, various railway objects are replaced by polygon meshes and rendered with a Phong shading model. Compared to the original point cloud using an octree, the results show a reduction of more than 95% in both memory usage and average rendering costs as well as an improvement in the discernibility between objects.