一种基于视频的人体动画骨架提取技术  

A VIDEO BASED HUMAN SKELETON EXTRACTION TECHNIQUE FOR ANIMATION

作　　者：庄越挺[1,2] 刘小明[1,2] 潘云鹤[1,2] 

机构地区：[1]浙江大学人工智能研究所,杭州310027 [2]浙江大学微软视觉感知联合实验室,杭州310027

出　　处：《计算机研究与发展》2000年第4期498-506,共9页Journal of Computer Research and Development

基　　金：国家自然科学基金

摘　　要：人体动画是计算机动画中最具有挑战性的领域 .针对传统人体动画技术的不足 ,提出了一种新技术来提取视频中的人体骨架 ,以应用于人体动画 .对于任何视频流 ,该方法都能在图像序列中跟踪人体特征骨架 ,并建立了透视投影下的三维人体运动骨架序列 ,最终通过自动注释运动信息建立了可供动画师浏览、查询的运动信息库 .这种方法具有素材来源丰富、计算量少、制作高效等特点 ,而且产生的人体运动非常真实 ,同时也将动画师从枯燥的工作中解放出来 .论文最后给出了实验结果 .Human animation is a challenging domain in computer animation. To overcome the shortcomings of conventional techniques, a new video\|based human skeleton extraction technique is proposed for animation application. Given a clip of video, human joints are tracked in image sequence, then the corresponding three\|dimension human motion skeleton sequence is constructed under the perspective projection, and finally a motion library is established by annotating multiform motion attributes automatically, which can be browsed and queried by the animator. This approach has the characteristics of rich source material, low computing cost, efficient production, and realistic animation result. The technique is demonstrated on several video clips of people doing full body movements, and the results in re\|animating a 3D human skeleton model are visualized.

关 键 词：人体动画 视频 卡尔曼滤波 人体骨架 计算机动画 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]