Video Automatic Image Stitching Research And Realization

Full-view panoramic image mosaics has become a important way to study virtual reality, computer vision. I have upgraded the traditional video surveillance systems using the technology to make it to be better. The system using six cameras with overlapping perspectives and video capture card collecting video stream on Linux operating system. Then features from the overlapping areas of adjacent video images were extracted and matched. Finally, a YUV format panoramic video to be encoded and played was generated . This paper presents a novel approach to creating full-view panoramic mosaics based on learning the traditional theory and technology. I have conducted in depth study in the following aspects. Full-view panoramic image mosaics can be divided into image registration and image blending. In the image registration, I use SIFT algorithm to extracting and matching features from continuous two images automatically and accurately. And then I use RANSAC algorithm to optimize the result of matching. In the image blending, in order to reduce the exposure differences and eliminate ghosting effects, I use the improved weighted average method and multi-resolution blending method. As high real-time requirements of the video image mosaic system, the feature extraction algorithms, matching algorithms and image blending algorithm has been optimized. Eventually I effectively achieve a panoramic video stream which can truly describe the scene of the world. The paper also compare the improved algorithm with the traditional methods to prove the feasibility of improving the algorithm.