Online Density Reduction Algorithm for Non-homogenous Multidimensional Datasets with Sequential Input

This paper introduces an online density reduction algorithm for non-homogenous multidimensional datasets. This work is motivated by applications in terrain modeling using laser data. Unlike most existing density reduction algorithms, the developed algorithm is designed for sequentially provided data and online implementation. Experimental results are presented for a 3D points cloud with 3 million points obtained via high resolution Riegl LMS Z620 scanner. The experimental results show that after processing the point cloud, the new density reduction algorithm uniformly reduced the number of points in the cloud from 3M down-to 48K points without decreasing the spatial representation quality of the scene. The algorithm presented here is shown to be efficient in terms of both memory and processer usage and and can be implemented for real-time applications.