Fast extraction of dominant planes in MLS-data of urban areas

Many current LIDAR systems generate huge amounts of 3D points, therefore efficient and fast data processing is essential. In urban areas surfaces are often planar and plane patches are an efficient representation for localization purposes. Our approach thus stores the points in a voxel grid and afterwards extracts the dominant plane using RANSAC-based plane fitting. Classical RANSAC-based plane extraction can result in improper planes, especially in the case of steps like curbsides. Different solutions have been presented, but they are computational demanding. We tackle this problem by extending the loss function of the standard RANSAC. This solves the problem in most cases and has nearly no impact on the runtime. We show the improvement using data of an urban environment, which were recorded by an MLS-system.