Video Image Stitching Technology

Video mosaic is a popular topic on video technology that shows significant important application for panoramic images, binocular robot and so on.Different techniques of video mosaic are used in different applications. In this paper, three types of case are studied:The method of phase correlation based on frequency domain is studied for a video sequence of translation. First, the principle of phase correlation and the algorithm process of video mosaic based on this method are introduced. Then, with a comparison of experiments, this method is suitable for a large overlap of video sequences between frames with certain small moving objects. A program of real-time stitching for video sequence with a small translation is designed.Two kinds of mosaic technology based on point feature are studied for two real-time videos synchronously captured from different viewpoints. First, the principle of Harris corner detector and SIFT operator are introduced. Then, a binocular video mosaic technology that employs point features is presented. The technology consists of several steps such as the matching of point pairs, the removal of false matches by use of RANSAC, the calculation of the model parameters in a perspective transformation matrix, interpolation, and pixel fusion. Concerning pixel fusion for stitching images, a method of automatic brightness adjustment and weighted fusion is adopted for eliminating stitching seams and unnatural color transition through experimental comparison. Due to the noise sensitivity of the Harris corner detector, in this paper, an improved method using voting mechanism is put forth for enhancing the accuracy of the alignment of the corners. Finally, real-time video mosaic programs for dual cameras respectively based on Harris corner detector and SIFT operator is implemented, in which the former is tolerant to image translation and rotation, while the later can resist translate, rotate and scale changes.At last, in this paper, mosaic of a sequence of images based on SIFT are studied. It is also applied to the mosaic of a sequence of text images which are captured by cell-phone.