2021/03/28 by Raza Yunus, Yanyan Li, Yunus, Raza +3 · 2 citations
Computer Science · Earth and Planetary Sciences · Engineering · #3D Surveying and Cultural Heritage #Advanced Vision and Imaging #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Robotics (cs.RO) #Robotics and Sensor-Based Localization
paper · pdf · doi:10.48550/arxiv.2103.15068
openalex publication_date 2021/03/28 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
In this paper, a robust RGB-D SLAM system is proposed to utilize the\nstructural information in indoor scenes, allowing for accurate tracking and\nefficient dense mapping on a CPU. Prior works have used the Manhattan World\n(MW) assumption to estimate low-drift camera pose, in turn limiting the\napplications of such systems. This paper, in contrast, proposes a novel\napproach delivering robust tracking in MW and non-MW environments. We check\northogonal relations between planes to directly detect Manhattan Frames,\nmodeling the scene as a Mixture of Manhattan Frames. For MW scenes, we decouple\npose estimation and provide a novel drift-free rotation estimation based on\nManhattan Frame observations. For translation estimation in MW scenes and full\ncamera pose estimation in non-MW scenes, we make use of point, line and plane\nfeatures for robust tracking in challenging scenes. %mapping Additionally, by\nexploiting plane features detected in each frame, we also propose an efficient\nsurfel-based dense mapping strategy, which divides each image into planar and\nnon-planar regions. Planar surfels are initialized directly from sparse planes\nin our map while non-planar surfels are built by extracting superpixels. We\nevaluate our method on public benchmarks for pose estimation, drift and\nreconstruction accuracy, achieving superior performance compared to other\nstate-of-the-art methods. We will open-source our code in the future.\n