The Study Of Panoramic Video Stitching Algorithm For Virtual Viewpoint

Nowadays, bulks of researchers have been enchanted by a new method in the area of computer vision, which is called as virtual close-up. Thanks to dedication of those elite, various algorithms of virtual close-up have been introduced in catena, one of the most certain popular methods among which is based on images, for its simplicity in image processing using simple image fusing to replace the traditional ways of geometry-based rendering. This paper makes a combination of this images-based virtual close-up and optical flow field effect to make offset the leakage in processing deep information of twodimensional images.Optical flow filed has been attached growing emphasis in image processing, holding different application to describe an accurate consequence of object motion and change, as well as robotic navigation, in which situation, this method expresses conspicuous advantages comparing others. Based on image drawing, this paper introduce optical flow filed, and then propose a new way of image processing which add optical flow field to virtual close-up. This method brings authentic consequences for virtual close-up, and can be transplanted to robotic navigation platform. At last, this paper adopts groups of natural images for training, and makes deep research in image fusing and image close-up strategies based on optical flow field. The work of this paper is as followings:Propose an image fusing method on blocking. This paper uses a combination of flow field to estimate image block, fusing strategy to eliminate image blocking, and weighted average to fuse images to obtain seamless fusing images, whilst descending time consumption for image processing.Propose a creative method of Realizing automatic generation for virtual images. Herein, the virtual image is generated automatically using progressive multi-step operation, namely: image preprocessing(color and geometric correction), the optical flow fielding, roaming between in the views, depth-class amplification. These constitute a completely autonomous generation system. This paper propose a new way that uses energy function to calculate the optical flow field between input images, the key of which is: firstly, using the traits of optical flow field and homograph in image series to estimate viewpoint change, and secondly, using optical flow field to calculate deep consequences in different image area, in order to the outline of the image foreground and background.Propose a method of simple robotic navigation based on virtual images. Robots perform simple navigation in completely unfamiliar surroundings by using image registration method and calculating the Euclidean distance between the images which are generated algorithmically. This method completely performed well without the help of other external tools, also does not need to create a map. It has some advantages such as a small data storage, lower computational complexity and so on.