带显著性区域约束的高效视频全景拼接方法  

An efficient panoramic stitching method with significant region constraint

作　　者：范菁[1] 吴佳敏[1] 叶阳[1] 吴冬艳[1] 王浩[1] 

机构地区：[1]浙江工业大学计算机科学与技术学院,浙江杭州310023

出　　处：《浙江工业大学学报》2015年第5期479-486,共8页Journal of Zhejiang University of Technology

基　　金：浙江省重大科技专项项目(2013C01112;2012C01SA160034);杭州市重大科技创新专项(20132011A16)

摘　　要：传统的基于流形的全景拼接技术在处理包含局部运动的视频时,存在不能保留特定关键区域的内容、表达视频运动信息的能力较差以及实现速度较慢的问题.针对上述问题,提出了一种带显著性区域约束的高效视频全景拼接方法,该方法在基于流形的视频拼接的基础上考虑图像的特征点,通过设定显著性区域,生成关键帧带约束的全景图;然后对关键帧全景图进行对齐及融合,构建运动全景图;并采用基于CUDA的并行计算方法进行算法加速.实验结果表明:该方法不仅可以在全景图中保留特定对象某一时刻的具体形态,而且可以保留运动对象多个运动形态,并且有较快的拼接速度.The traditional panorama stitching technique based on manifold cant retain the contents of particular key areas in videos.The ability to express the motion information of video is poor and the speed is slow.In order to solve these problems,an efficient panoramic stitching method with significant region constraint is presented.Based on the manifold video stitching and the image feature points,this method will set significant area to generate the panorama with constraint for key frames firstly.Then the panoramas in the key frames are aligned and fused to obtain the motion panorama. The CUDA parallel computing is used to speed up. The experimental results showed that the proposed method can not only retain the specific form of specific object at a certain moment,but also can keep multiple movement patterns of moving objects in the panorama.It can effectively improve the speed of image stitching.

关 键 词：视频拼接 流形 特征点 全景图 对象约束 CUDA技术 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]