LARGE-SCALE SURFACE RECONSTRUCTION THAT IS ROBUST AGAINST TRACKING AND MAPPING ERRORS

摘要

A depth map of a physical space is obtained using a depth sensor mounted on a rig, for example, a head mounted display device worn by a robot or a user. A visible light image is also obtained. Image and orientation readings can be used to create a pose graph comprising nodes connected by links. The nodes are associated with different poses and corresponding images of the rig. Links between nodes represent a correspondence relationship between images and a transformation relationship between node coordinate systems. New images are captured and the pose graph updated to reduce error accumulation. In addition, the surface in the physical space can be reconstructed at any time according to the current state of the pose graph. The volume used in the surface reconstruction process is fixed to the node, so the position of the volume is adjusted when the pose graph is updated.