Research On Image Stitching And Wide Angle Panorama Construction

Panorama auto-generate techniques play an important role in Image Based Rendering (IBR) technology, which refers to computer vision, image processing and computational geometry.Most panorama auto-generate algorithms follow three steps listed below: image capturing, image matching and image reprojection.In the image capturing step, this paper made a brief introduction to camera movements and the corresponding image transformations. The transformations caused by camera movements can be regard as perspective projection transformation which has 8 degrees of freedom (dof) in its parameter. With these 8 dof, a transformation can be distinctly identified.A simple and efficient image feature detector named Harris corner detector is used in image matching step. Some improvements are made to this algorithm. A feature descriptor is formed for each feature point, and is used in the following feature matching works. In feature matching step, a highly efficient search algorithm based on indexing of Harr wavelet coefficient is brought forward in this paper. Then, a pseudo matching set is obtained by comparing local neighborhoods of features and these pseudo matches are divided into inliers and outliers using robust RANSAC algorithm. With the inliers, an initial transformation parameter can be obtained.A panoramic connected sequence can be found in image reprojection step. The connected images can form sequence trees. Images are drawn one by one as traverse the tree nodes, and whose parameter can be incidentally optimized by LM algorithm. Finally, three different reprojection methods can be applied to the images: wide angle planar reprojection, 360 degree cylindrical reprojection and full-view spherical reprojection.According to the algorithms above, a highly useful panorama auto-generate system PanoProducer is developed.