Research Of Panorama Stitching Technology Based On Video Stream Captured By Web-camera

Image mosaicing technology is a hot research topic in computer vision, computergraphics and virtual reality. Image mosaicing has a wide range of applications in such areasas virtual reality, video conference, scene monitoring, medical microscopic video and so on.At present, we can obtain panoramic images directly by ultra-wide angle lens and fisheye lens.But these devices are expensive and complex, and the images obtained by these devices havedistortion. By studying panorama stitching techniques, this paper takes video sequencesobtained by rotating web-cameras as the data source. And based on such techniques as keyframes selection, feature extraction, feature matching and image fusion, a panorama stitchingsystem based on video stream captured by web-camera is designed and developed.The main topics of this research are as follows:(1) We have analyzed and studied the two most popular feature extraction algorithms, i.e.SIFT and SURF. Both algorithms have good noise immunity and robustness, and the SURFalgorithm is much faster than SIFT algorithm. In this study, we adopt SURF algorithm toextract characteristic points, and to the problem that the feature points detected from videoimage by SURF algorithm are numerous and dense which is not conducive to matching andcalculating transformation matrix, a optimized SURF algorithm is presented. According tothe experimental results, the optimized SURF algorithm works faster than the original SURFalgorithm, and the features extracted is distributed evenly.(2) We have employed the Nearest Neighbor matching algorithm and BBF searchingstrategies for feature matching during image matching. And then RANSAC algorithm whichhas geometric constraint is used to eliminate wrong matching and estimate the matrix oftransformation relationships. The experimental results show that this method can obtain moreaccurate matching result and transformation matrix.(3) We have done experiments to compare the effects of evolutional fusion algorithmand multi-resolution fusion algorithm in eliminating the obvious mosaicing seam. Theevolutional fusion algorithm works faster than multi-resolution fusion algorithm, but the stitching result has fuzzy because of small geometry dislocation. The multi-resolution fusionalgorithm can retain more detailed image information, the fusion quality is higher than that ofevolutional fusion algorithm. So we use multi-resolution fusion algorithm in this paper.(4) We have designed and developed a panorama stitching system based on video streamcaptured by web-camera. The video sequences captured from different environments areanalyzed by experiments. The experimental results show that this system can realizepanorama stitching based on video sequences captured from complex environments.