Research On Augmented Reality 3D Registration Technology Based On Vision

Augmented reality is a research hotspot in computer vision in recent years.It has been applied to various fields,and virtual wear is a very important application direction.For augmented reality systems,3D registration is the core technology.The 3D registration technology estimates the position of the camera based on the known 3D and 2D matching information,and obtains the relative positional relationship between the virtual object and the real object,thereby superimposes the virtual object into the real scene.This paper aims to studying the vision-based 3D registration technology,proposing methods of camera pose estimation,and establishing an augmented reality system.Most of the existing three-dimensional registrations estimate the camera pose by placing artificial markers,which limits the application range of augmented reality,so this paper uses natural features for pose estimation.How to quickly and accurately estimating the camera pose is studied with the 3D and 2D matching information.The main research work of this paper is as follows:(1)A pose estimation method based on known camera internal parameters is proposed.Under the premise that the camera internal parameters are known,the problem is: Knowing the 3D-2D matching point pairs,how to estimate the camera pose.Aiming at the problem that the camera pose estimation accuracy is not high,the virtual control point is constructed,and the POSIT algorithm is combined to obtain a more accurate initial pose,thus improving the accuracy of the estimation result.Aiming at the problem that the camera pose is slow,the initial position of the camera is calculated from the coordinate information of the virtual control point,which simplifies the calculation process of the EPnP algorithm,thereby reducing the calculation time.The experimental results show that the proposed method has faster calculation speed,higher precision and stronger robustness than the existing methods.(2)A method for estimating the focal length and pose of an uncalibrated camera is proposed.In the case where the camera internal parameters are unknown,the problem is: Knowing the 3D-2D matching point pairs,how to simultaneously estimate the camera focal length and pose.The method proposed in this paper adds the unknown variable of the camera focal length based on the estimated pose.The method calculates the initial value of the focal length by the relative distance between the virtual control points,and optimizes the focal length and posture by using the distance of the virtual control points in different coordinate systems.The experimental results show that the proposed method has faster speed and higher precision than the existing methods,and has stronger anti-noise ability.Under the premise of ensuring the accuracy of pose estimation,the camera focal length can be accurately estimated.(3)Design and implement an augmented reality system based on 3D registration.The system uses the camera pose estimation method proposed in this paper to design the virtual glasses try-on scene.The system has five modules: the scene acquisition module uses the camera to obtain the video image of the real scene;the three-dimensional registration module estimates the pose of the camera relative to the face;the model generation module creates a virtual three-dimensional glasses model;the virtual and real fusion module superimposes the glasses model Corresponding position;the display module outputs the result of the virtual fusion.The system uses only one monocular camera,which is low in cost and easy to set up.It can be tried in real time.The virtual glasses can follow the movement of the face in real time.The pose of glasses can be consistent with the pose of the face,and the virtual glasses try-on result is vary realistic.