Research On Real-time Video Mosaic

Video mosaic is widely used in the fields of video conferencing, videosurveillance and so on; it can increase the field of view while not reducing resolution.Due to large amount of data, complex environment and other reasons，real-time videomosaic becomes an active area of research. In this paper, technique of real-time videomosaic is studied.The core of video mosaic is image mosaic. Firstly, the general flow of imagemosaic and the definition of image registration are introduced and the commontechniques of image registration are analyzed. Then the image mosaic based onfeature extraction, matching and image fusion is adopted in this article.Then the approach SURF which has good performance is analyzed and improved,so that it can be utilized to real-time video mosaic. Improved SURF limit the scope ofthe feature extraction and reduce the dimension of feature descriptors. After extractthe feature points with the method of improved SURF, it comes to the matching stage.The method of ratio of the closest neighbor and the next-closest neighbor is used forcoarse matching and RANSAC is adopted to purify the matching pairs and calculatethe transformation matrix. Experimental results demonstrate that the improved SURFoutperforms original SURF with respect to speed of feature extraction and featurematching and at the same time robust to rotation, scale changes and change inillumination conditions. Then some simple image fusion methods are compared andgradating in-and-out algorithm is chosen because it can eliminate the seam andachieve a smooth transition of the image.Finally, a video mosaic platform based on USB cameras is achieved. We calculatethe transformation matrix for a period of time and use the calculated transformationmatrix to stitch the other frames. Experiments show that we can achieve video mosaicwith high speed by using our method.