Research And Implementation Of Image Stitching Based On Video Image

With the development of digital image technology, image stitching technology has been become hot points research on the computer vision, image processing and computer graphics science, which is widely implicated in the fields of virtual reality, medical image processing, remote sensing and military. As the video is limited by the camera vision, it can not directly obtain the wide angle image. Image stitching is a technology for forming a wide view image, which can address the limitations of image vision. This paper focuses on the research of the video image stitching technology.According to the process of image stitching, this paper firstly introduces the research background, the significance of researching, the current researching situation and core technologies of image stitching. Then, it summarizes and epitomizes the relevant theories of image stitching. The paper focuses on the research of the image feature extraction, image registration and image fusion. In feature extraction, an improved method of Harris corner detection is presented based on the traditional method. It suppress the corner and filter the pseudo corner using the conditions and response matrix maximum. It effectively solves the problem of pseudo feature points of traditional Harris algorithm. In feature matching, it improves the traditional pyramid Lucas-kanade light flow algorithm and the corner matching algorithm in this paper. It realizes the effective corner matching and filter the moving corner of the video by two-way track for the light flow of Lucas-Kanade. In image registration, affine transformation model is not robust, which can not apply in violent jitter videos. Therefore, the motion compensation of the camera model bases on self-motion parameters is proposed, which solves the image registration of violent jitter videos. Finally it achieves the image fusion taking the improved synthetic method of gradually progressive.Experiment results show that the proposed stitching algorithm of video images still performs well in the case of rotation, translation and noise distractions. And it has a certain robust and can meet practical demands.