Real-time adaptive point splatting for noisy point clouds

Regular point splatting methods have a lot of problems when used on noisy data from stereo algorithms. Just a few unfiltered outliers, depth discontinuities, and holes can destroy the whole rendered image. We present a newmulti-pass splatting method on GPU hardware called Adaptive Point Splatting (APS) to render noisy point clouds. By taking advantage of image processing algorithms on the GPU, APS dynamically fills holes and reduces depth discontinuities without loss of image sharpness. Since APS does not require any preprocessing on the CPU and does all its work on the GPU, it works in real-time with linear complexity in the number of points in the scene. We show experimental results on Teleimmersion stereo data produced by approximately