Consolidation of multiple depth maps

Consolidation of point clouds, including denoising, outlier removal and normal estimation, is an important pre-processing step for surface reconstruction techniques. We present a consolidation framework specialized on point clouds created by multiple frames of a depth camera. An adaptive view-dependent locally optimal projection operator denoises multiple depth maps while keeping their structure in two-dimensional grids. Depth cameras produce a systematic variation of noise scales along the depth axis. Adapting to different noise scales allows to remove noise in the point cloud and preserve well-defined details at the same time. Our framework provides additional consolidation steps for depth maps like normal estimation and outlier removal. We show how knowledge about the distribution of noise in the input data can be effectively used for improving point clouds.