Rendering of Dense, Point Cloud Data in a High Fidelity Driving Simulator

Abstract

Driving Simulators are advanced tools that can address many research questions in transportation. Recently they have been used to advance the practice of transportation engineering, specifically signs, signals, pavement markings, and most powerfully to examine the safety and efficiency of alternative transportation solutions. These simulators are a powerful 3D, virtual environment enabling the study of how drivers respond to potential designs or policies. A key challenge is virtual environment that maintains high fidelity to the real world. 3D laser scanners, which use Light Detection and Ranging (LIDAR), are line-of-sight technology that emits laser pulses at defined, horizontal and vertical angular increments to produce a 3D point cloud, containing XYZ coordinates for objects that return a portion of the light pulse within range of the scanner. This detailed point cloud is a virtual world that can be explored and analyzed by a variety of people. Through the combination of these two technological systems, more authentic, virtual, built-environments can be used by transportation engineering professionals for the purpose of 3D design. This research project focuses on the technical issues of importing and displaying 3D laser scan data within a driving simulator. For import in the simulator, datasets need to be in the VRML97 format with color values scaled from [0 1]. A transformation needs to be applied to convert between real-world coordinates and screen coordinates. Large datasets should be filtered, when possible, and tiled into very small increments (< 35 MB) to maintain system interactivity.