Using geometric primitives to render live RGB-D data in the Occulus Rift

With advances in technology and lowered price the use of RGB-D cameras for robot applications has become popular. They are able to provide rich information about the environment but due to the huge amount of data that they produce and often limited computational resources, processing and analysing the data is challenging. This creates the need for good and efficient compression methods. In this thesis we suggest a lossy compression method that extracts planar surfaces from point cloud data, removes redundant interior points and stores the planes as a triangulation of the remaining points. The method can remove over 95% of input points for a given plane and can do so in real time making it suitable for robotics applications. Despite high compression ratio, the resulting compressed point cloud stays true to the original scene and is visually pleasing to look at.