Point Clouds

First demonstration of an interactive 3D fly-over of a city. The data is a point-cloud originating from an Aerial LIDAR scan. This particular dataset is a part of the city of Middelburg, made available in the AHN2 test data set [2]. Colors are reconstructed from 2D Texture, so these are not the directly scanned colors.

We generate a level-of-detail data structure in OpenSceneGraph to dynamically show different densities of points, and is loading over the network. With this ~20 Million point cloud, update rates are between 15~60 fps depending on how many points are visible in a certain view. We are optimizing the datastructure to balance frame rate with visual quality. 3D Navigation is done with a SpaceMouse device.

Point Cloud visualizations have nice visual quality, especially under motion and 3D stereo, both of which are used in the VR systems in our group [1].
Please note that these visualizations are inherently difficult for regular video compression algorithms to process without introducing disturbing visual artifacts. Download the wmv file below for a higher-quality impression, or see a static screenshot [3].

[1] visualisation.tudelft.nl/VRVis
[2] ahn.nl/
[3] graphics.tudelft.nl/~gerwin/media/3dtopo/Screenshot_pointcloud.png

Gerwin de Haan