Feature-aware LiDAR point cloud simplification (POSTER)

Ma et al. (2012) introduced the shrinking ball algorithm to approximate a point approximation of the MAT from an oriented input point cloud.For each sample point, we start with a very large tangent ball that is centered along the point's normal.An empty maximal tangent ball is found by iteratively reducing the ball's radius using nearest neighbor queries from its center point.Compared to earlier algorithms (e.g.Amenta et al., 2001) it is simple, fast, robust in practice, and easy to parallelize.This makes it a good choice for approximating the MAT of large LiDAR point clouds. Shrinking ballsThe Medial Axis Transform (MAT) is defined as the set of maximal balls tangent to the object surface at two or more points.The centers of these ball form the Medial Axis, a medial skeletal structure.LiDAR point clouds are highly detailed and cover large areas.This brings great advantages for applications such as flood modeling, crisis management and 3D city modeling.Unfortunately, and despite recent developments on this subject, current methods from practice are unable to fully take advantage of modern LiDAR datasets.First, because of their huge data volume they do not fit in a computer's internal memory.As a result, many of the conventional software tools have become very inefficient.And second, many existing methods use only 2.5D data-structures and algorithms.While this alleviates memory requirements and simplifies computation, it comes at the price of a significant loss of information, because valuable 3D information that is present in LiDAR point clouds is ignored.