Panorama-based Camera Pose Estimation

Nowadays,spherical panorama has become a common data form of scene recording and visualization.Panoramic spherical image presents a 360-degree scene captured at a single point in space,which can be produced by omnidirectional camera shooting or by combining multiple plane perspective images.On the other hand,structure-frommotion technology for plane perspective images is becoming more and more mature.This technology uses multiple two-dimensional perspective images containing visual motion information to estimate the three-dimensional structure of the shooting scene.However,due to the geometric characteristics of spherical panoramic images,the existing technology of motion recovery structure cannot be directly applied to spherical panoramic images,and the relevant technology of spherical panoramic images is still in the early stage of research,so it is urgent to carry out targeted research.Through the study and analysis of the epipolar geometry relationship between panoramic spherical images,this paper develops and implements a method of motion restoration for spherical panoramic images.In this paper,an effective parallel computing technique of feature matching is proposed to improve the speed of feature point matching.First,matching points are selected from all the images to initialize the two images with the largest number,and the relative camera pose of the initial structure is estimated by combining the recovered 3d structure.Then select images by selecting strategy,add an image and estimate the camera pose of the new view by using the two-dimensioned and three-dimensional correspondence.In the optimization stage,global optimization and local optimization are adjusted alternately,and re-triangulation is added to reduce the estimated error as much as possible.Finally,in a visual system interface,the relative camera pose of several spherical panoramic images and the sparse point cloud of the restored 3D scene are displayed.Through comparison with the relative camera pose of the artificially set analog camera,the experiment shows that the results obtained by the method in this paper are relatively accurate.