基于自适应帧采样的视频拼接  

Video mosaicking based on adaptive sampling

作　　者：刘永[1] 王贵锦[1] 姚安邦[1] 汪浩 林行刚[1] 

机构地区：[1]清华大学电子工程系,北京100084 [2]诺基亚中国研究院,北京100176

出　　处：《清华大学学报（自然科学版）》2010年第1期108-112,共5页Journal of Tsinghua University(Science and Technology)

摘　　要：针对视频拼接计算量过大的问题,提出了一种分层式自适应帧采样的视频拼接方法。两层帧采样环节分别采用了不同的策略:第1层建立了帧间重叠率和帧间隔的分段线性模型,通过"预测-检验-修正"环节不断的更新采样帧间隔来自适应抽取到满足一定重叠率的准关键帧;第2层在第1层采样的基础上,利用逐帧检测的方法从准关键帧中抽取到满足一定条件的最终关键帧。最后对关键帧进行全景图拼接。与用所有视频帧进行拼接的方法相比,该方法用极少的场景信息损失换取了拼接效率的大幅提高。实验结果表明:该方法对于一般实验场景均能鲁棒地抽取到可靠关键帧,在大幅度降低计算负荷的同时得到高质量的视频全景图。A hierarchied adaptive frame-down-sampling based scheme was developed to reduce the video mosaicking computation cost. Different strategists are used in the hierarchical frame-down-sampling. At the first sampling stage, dynamic piecewise linear model is constructed based on the inter-frame interval and overlap latio. Estimating-checking-modifying loop is adopted in this mode to adapt the frame-sampling interval and select the candidate key frames. The second stage checks the frame-to-frame overlap ratio among the candidate key frames and decides the final key trames. Finally the panorama is constructed from the sampling key frames. Compared with the traditional video mosaicking which utidzed all the frames, this scheme is much more efficient without losing essential visual information. Experiments on real video sequences thow that this algorithm can extract robust key frames and achiew high-quality video panoramas with low computational costs.

关 键 词：视频拼接 图像配准 关键帧提取 自适应帧采样 全景圈 

分 类 号：TP591[自动化与计算机技术]