摘要
运动捕获数据行为分割的目的是将长序列数据划分为单个运动类型短片段的序列集合,使集合中每个片段具有特定的运动语义。针对相邻运动片段的过渡区间存在部分运动帧序列的语义归属歧义,提出了一种结合双特征的运动捕获数据行为分割方法。该方法首先从原始数据中提取角度和距离两组不同类型的运动特征集,并分别基于PPCA方法构建规格化的综合特征函数;然后利用子区间标准差阈值限定方法分别对综合特征函数进行粗分割,从而将运动捕获数据划分为若干具有独立语义特性的可信区域与待定区域;最后采用高斯混合模型方法判别待定区域的具体归属,从而得到最终的分割结果。实验结果表明,该算法能对模糊歧义区域进行细分割,具有较好的分割效果。
The objective of motion capture data behavior segmentation is to divide the original long motion sequence into several motion fragments, and each motion fragment incorporates a particular semantic behavior. In general, the transi- tion parts of some neighboring motion fragments are always encountered with the semantic ambiguity. To this end, this paper presented a double-feature combination based approach to ＇tackle this problem. The proposed approach first ex- tracts two different types of motion features, i. e. , angle, distance, and then utilizes the PPCA algorithm to construct two different comprehensive characteristic functions individually. Subsequently, a subinterval standard deviation ap- proach associated with threshold limiting strategy is employed to segment the comprehensive characteristic functions in- to several confidence regions and pending regions roughly. Finally, by utilizing the Gaussian mixture model to further determine the pending regions, the robust segmentation result can be obtained. The experimental results show that the proposed approach performs favorably compared to the state-of-the-art methods.
出处
《计算机科学》
CSCD
北大核心
2013年第8期303-308,共6页
Computer Science
基金
国家自然科学基金项目(61202298
61202297
61102163)资助
关键词
运动行为分割
双特征
综合特征函数
可信区域
待定区域
Motion behavior segmentation
Double feature
Comprehensive characteristic functions
Confidence region
Pending region
