The Large FOV Image Mosaicing

In the field of computer vision,image stitching has always been a hot research direction.It can be used to generate large-field and high-resolution images to meet people's needs.With the development of related technologies,image stitching has now been widely used in virtual reality,video monitoring,medical image processing,military reconnaissance,remote sensing and many other fields.As the basis of this paper,the method of image registration characterized by point feature is given the most attention.Started from minimizing the registration error,this paper,under the guidance of related theories based on different models,has studied control points sampling and further discussed that the uniform sampling of control points is crucial to accurate registration.Then,the related problems of image stitching in static platform have been studied in this paper.As the first step,Zhang's method is used to do the camera calibration job,and correct the captured images according to the calculated camera parameters.Then calibration paper is applied to complete the image projection and registration work,and explain the reason why all the images are projected to the flat and why the nearest neighbor interpolation method is chosen.Finally,according to the practical application requirements,a video stitching algorithm of fixed cameras based on control images is proposed in this paper.All the parameters related to panoramic image are calculated during the initialization,and the look up table has been applied to project each image captured from the cameras to the panorama image.In this way,It can skip the complex calculation process to achieve real-time results.At last,in order to get large field of view images in moving platform,,sub-block feature detection is presented,based on the conclusion made before.After feature extraction,BBF method is used to find out matched features between reference image and the image to be registrated,and transformation parameters are robustly calculated by using RANSAC.In the end,some ghost-removing methods are proposed to handle the situation of fusion ghosts which may occur in the panoramic image,including multi-band blending,seam-line searching and motion area detecting,which are proved effective through the experiment.