体育视频全景图合成技术

A Panorama Composition Technique of Sports Video

研究了面向体育视频的全景图合成技术,提出了通过计算前景区域模板和α混合进行前景合成的方法.首先采用MPEG-4校验模型中的Konrad算法和基于Fisher线性判别准则的外点过滤算法进行全局运动参数估计,并用中值法合成背景全景图;然后将前景帧进行仿射投影后与背景求差值,并用中值滤波和形态学方法处理差值图像,得到前景区域模板;最后采用α混合生成前景序列的全景图,并叠加到背景上.由此得到的全景图能同时展现运动的场景和运动员的动作过程.

A technique of composing mosaics from sports videos is proposed. First we select means of the temporal median for background composition, then we propose a foreground mix technique based on foreground region extraction and α-composition. To estimate the global motion, Konrad algorithm investigated in the framework of MPEG-4 and an outlier filter based on Fisher discriminant are used. The foreground mask is extracted from margin image by median filter and morphological filter. We add foreground region to background with α-composition. The final panoramas clearly show the full background and vivid actions of athletes.