Research On Monocular Vision Multi-user Augmented Reality Based On Point Cloud Registration

Augmented reality is a technology that superimposes computer-generated visual data into the real-world scenes and provides interactive functions to its users.The multi-user augmented reality mechanism is a mechanism that allows multiple users to interact with each other in the same real-world scene.By solving the coordinate transformation between different devices,the virtual objects which are put in augmented reality space can be shared.There is a problem with the existing multi-vision mechanism that the equipment is so expensive that it cannot be popularized.There is a problem with the mechanism based on visual inertial odometry that it has poor versatility caused by the dependence on the strictly calibrated IMU with high precision.In the existing monocular vision mechanism,there is a problem caused by its dependence on markers.In addition,although monocular vision SLAM can be used instead of the marker to provide three-dimensional registration for augmented reality and make it more general,the monocular vision SLAM cannot determine the true scale of the real-world scene,which will bring the problem that the coordinate systems between different devices cannot be unified.Aiming to solve the above problems,this thesis propose a monocular vision multi-user augmented reality mechanism based on point cloud registration.The major work of this thesis is divided into three parts:(1)A point cloud registration method of different scale based on deep neural network is proposed.This method uses deep neural network to extract the spatial features of the point clouds,and then performs iterative registration based on the extracted spatial features.By introducing a scale factor and similarity transformation with its corresponding 7-dimensional Lie algebras,this method acquires the ability of registration of point clouds with different scales.(2)Based on the proposed point cloud registration methods,a monocular vision multi-user augmented reality mechanism is proposed.This mechanism uses monocular vision SLAM to provide three-dimensional registration capabilities,which has high versatility and can avoid the dependence on markers.The coordinate transformation module uses the proposed method to unify the coordinate systems between different devices.(3)The effectiveness of the proposed method and mechanism is verified through experiments.The proposed method has been compared with the existing point cloud registration method.This thesis also create an experimental prototype of the proposed mechanism.The prototype has been functionally tested and compared with the existing multi-user augmented reality mechanism.The results show that the proposed method and mechanism have met the expectation.