Real-time CAD-level surface of revolution reconstruction on visual slam

Intelligent robots need to understand the geometry and semantic information of the surrounding environment in order to interact meaningfully with the scene. The use of shape constraints and semantic information in visual SLAM(Simultaneous Localization and Mapping) has proved to be a promising research direction. In this paper, combining the shape and semantic information of revolving bodies, CAD-Level semantic scene map is calculated in SOR region, which is ready for robot operation. The results model of the gyroscopic structure are independent of the sparsity of the point cloud map, and can be used for feature missing, highlights, and scenes which containing transparent revolving bodies.