Explorations for real-time point cloud rendering of natural scenes in virtual reality
Daniele Bonatto, Ségolène Rogge, Arnaud Schenkel, Rudy Ercek, Gauthier Lafruit · 2016
This work is a proof of concept, where we explore the possibility of rendering natural scenes in a head mounted display device without meshing. Real-time, stereoscopic full-HD rendering is obtained for a 14 million points scene, using a low end graphics card for virtual reality (Nvidia GeForce GTX 970), within an Oculus Rift DK2. High quality is achieved by using splatting, while real-time rendering is made possible by the means of a good data structure and a complexity reduction of the scene with techniques such as Optimized Sub-Sampling, Level of Detail and Frustum Culling. Altogether, those techniques lead to a good virtual reality immersion. Choices and limitations of the proposed techniques are discussed.