Design And Implementation Of Augmented Reality Oriented Camera Pose Estimation System

Augmented Reality is one of the branches of Computer Vision,it can brings user a composite visual experience,and is widely used in all walks of life.3D registration is the core of augmented reality.By calculating the relative position of the camera pose and vitual information,then add the vitual information accurately.In visual-based 3D registion,feature selection and matching is one of the most important part in camera pose estimation.Points are the most commonly used feature in feature selection and matching.However,in some cases,the redundancy of points will bring about the problem of large computational complexity.In order to reduce the complexity of pose estimation,the work of this paper are as follows:1.This paper mainly studied the selection of point features in pose estimation algorithm.The author aims to propose a pose estimation algorithm that based on information entropy theory when the sensor is monocular or stereo camera.After feature selection and matching,the algorithm calculates the mutual information between camera pose and 3D points,and retain the 3D points that meet the predetermined threshold and generate mappoints,use EPn P algorithm to estimate camera pose.Then use bundle adjustment and loop closing to optimize the camera pose.2.This paper implements a system: Mutual Information Augmented Reality.The system consists of pose estimation module,AR display module and other modules,and can be applied to real scenes.Tests shows that this system can run in real-time and meet the needs of computational complexity and accuracy.In this paper,points are selected to reduce the computational complexity of the camera pose estimation while ensuring the accuracy of the pose estimation.Testsshows that the algorithm is designed reasonable and can work well when the sensor is monocular or stereo camera.Therefore,the system is pratical.