Comparisons of Eight Simplification Methods for Data Reduction of Terrain Point Cloud

In recent years, the applications of 3D point cloud data representing terrain surfaces have been growing rapidly. Such data typically have a very fine spatial resolution, which can lead to computational and visualisation issues. To overcome these issues, it is a common practice to reduce the density of point cloud data during initial data processing. As such, various simplification methods had been developed and used in practice. The choice of those methods is crucial to preserve features and shapes of the terrain in the simplified point cloud data. Previous studies on this matter were focused mainly on the methods commonly used in geosciences, but did not consider those in computer graphics. In this study, a total of eight simplification methods that are used widely in both geosciences and computer graphics were compared and analyzed using four sets of terrain surface point cloud data. In addition, unlike previous studies where a global RMSE (root mean squared error) was used as the metric for comparing different methods, the standard deviation of local RMSEs (root mean squared errors) was also calculated in this study to check the uniformity of local RMSEs over the whole terrain areas considered. The results show that the adaptive sampling method yielded thinned point cloud data of higher overall accuracy and more consistent local RMSEs than those obtained using the other methods considered.