2018/04/24 by Markus Ylimäki, Ylimäki, Markus, Juho Kannala +3
Computer Science · Engineering · #Advanced Vision and Imaging #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Optical measurement and interference techniques #Robotics and Sensor-Based Localization
paper · pdf · doi:10.48550/arxiv.1804.08912
openalex publication_date 2018/04/24 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Depth map fusion is an essential part in both stereo and RGB-D based 3-D\nreconstruction pipelines. Whether produced with a passive stereo reconstruction\nor using an active depth sensor, such as Microsoft Kinect, the depth maps have\nnoise and may have poor initial registration. In this paper, we introduce a\nmethod which is capable of handling outliers, and especially, even significant\nregistration errors. The proposed method first fuses a sequence of depth maps\ninto a single non-redundant point cloud so that the redundant points are merged\ntogether by giving more weight to more certain measurements. Then, the original\ndepth maps are re-registered to the fused point cloud to refine the original\ncamera extrinsic parameters. The fusion is then performed again with the\nrefined extrinsic parameters. This procedure is repeated until the result is\nsatisfying or no significant changes happen between iterations. The method is\nrobust to outliers and erroneous depth measurements as well as even significant\ndepth map registration errors due to inaccurate initial camera poses.\n