摘要
基于虚拟立体视假设,借鉴RANSAC技术的思想,文中针对数据缺失(帧对之间匹配特征可能较少)情况下的视角无关手语识别问题,提出一种Sample-Consensus方法.其基本出发点是,同一手语不同视角下的两个样本序列之间所有的对应帧对,可以解释为由某一虚拟立体视觉系统同步捕获,因而满足同一个基础矩阵,而且此基础矩阵能够基于部分对应帧对包含的点对应关系进行估计.实验表明,提出的Sample-Consensus方法能够有效地应用于数据缺失情况下的视角无关手语识别.另外,这种方法也可以扩展到相近的领域,如视角无关的动作识别和刚体运动分析等.
This paper proposes a Sample-Consensus method for viewpoint independent sign language recognition under data deficiency （matched features are possibly deficient with regard to some frame pairs）. The proposed method is based on the epipolar geometry and inspired by RANSAC. The basic idea is that all corresponded frames between two sequences of the same sign can be roughly considered as captured synchronously by a virtual stereo vision system and thus they will satisfy the same fundamental matrix. In addition, the fundamental matrix can be estimated from point correspondences contained by some part of corresponding frames. Experimental results demonstrate the efficiency of the proposed method. Moreover, this Sample-Consensus method can be easily extended to some similar problems, such as viewpoint independent activity analysis and rigid-motion analysis.
出处
《计算机学报》
EI
CSCD
北大核心
2009年第5期953-961,共9页
Chinese Journal of Computers
基金
国家自然科学基金(60533030
60603023
U0835005)
多媒体与智能软件技术北京市重点实验室开放课题资助~~
