An approach to visualization of large data sets from LIDAR

Rapid development of laser scanning technology in past decades has resulted in a wide area of its applications. LIDAR is a system that uses this technology to gather information about distant targets. Gathered data are stored into large data sets that are further processed, visualized and analyzed. Fast and accurate visualization is the key factor when working with LIDAR point clouds. The main problem that arises is that vast amount of data can easily exceed memory and processing capacities of modern day computers. In this paper we present an approach to visualization of large LIDAR point clouds in real time entirely on graphical processing unit using a point-based rendering technique. Our method is based on dynamic data loading and efficient two-pass rendering utilizing approximation of elliptical weighted average splatting with rotated splats. Expensive rendering tasks are delegated to programmable graphics unit to save CPU resources. The proposed system offers realistic visualization of LIDAR point clouds in real time that is visually and performance wise comparable to other solutions, while not requiring any comprehensive preprocessing such as TIN generation beforehand.