Research And Realization Of Registration Method For Augmented Reality System

By supplementing or adding virtual objects and related information on a real-worldsense, make the virtual objects becoming parts of the around real world. Registrationtechnology has always been the focus and one of the most critical technologies, it’s ofgreat significance to implement Augmented Reality.The traditional visual registration technology is applicable in a relatively simplereal environment, such as the situation of the camera relative to the object stationary,once the scene is becoming complexity or the camera relative to the object movingrapidly, the computation of the registration algorithm will be a substantial increase, andthe result of the registration will be significantly worse, even fails in registration in thecase of some of the objects is blocked.In this paper, use the identity-based visual theory to implement the traditionalvisual registration technology, optimizing the processing of the three-dimensionalmatrix parameter equation, make only four sign points needed to complete the solvingof the registration matrix. Meanwhile, in order to solve the problem of instability in thecase of partial occlusion leading to registration failure, improved the calculation of3Dregistration matrix by using the homography matrix, designed to calculation the matrixparameters through the adjacent frames when the identity is part of blocked, the specificmethod is when the identity is fully visible, through the traditional visual algorithm tocomplete the three-dimensional matrix parameters calculation; if the identity is not fullyvisible, then through the geometry of four or more than four non-collinear feature pointsin the two adjacent images matched each other and the singular value decomposition, tocalculate the homography in the iterative procedure. Accordingly, decomposing thehomography matrix into unknown parameters of3D registration matrix to calculate theregistration matrix of the virtual registration required, then to complement the trackingregistration when the identity is not fully visible. The feature point’s matching using theoptical flow feature matching algorithm, aimed at identifying the non-collinear featurepoints of special identity in the real scene to complement the feature points matchingtrack of the adjacent images, so that can get the geometry information which is used tocalculate the homography matrix and registration matrix, virtual objects will eventuallyaccurately superimposed on the real scene.The innovation of this paper is the introduction of the homography matrix torecover the correspondence between multiple images, improved the3D registrationmatrix’s solving problem when the identity is partly blocked, in the same time, usingthe Hessian matrix in the optical flow matching algorithm of the registrationtechnology’s identification portion, so that the registration’s stability has beenimproved.The experimental results show that using the method of the homography matrixand the optical flow feature tracking technology extends the range of application of the technology of visual registration, ensures the registration of the virtual objects into thereal scene, improves the system’s stability, solving the registration failure problem whenthe identification is partly blocked. The registration algorithm in this paper reflects itseffectiveness in the case of the camera and the object is relatively fast-moving and theobject is partially blocked, it has some practical significance for three-dimensionalvirtual objects’ registration in the mobile Augmented Reality.