结合金字塔模型和随机森林的运动捕获序列语义标注  

Motion Capture Sequence Semantic Annotation Via Pyramid Model and Random Forests

作　　者：徐濛 彭淑娟[1] 柳欣[1] XU Meng;PENG Shujuan;LIU Xin(College of Computer Science and Technology, Huaqiao University, Xiamen 361021, China)

机构地区：[1]华侨大学计算机科学与技术学院,福建厦门361021

出　　处：《华侨大学学报（自然科学版）》2017年第6期848-853,共6页Journal of Huaqiao University(Natural Science)

基　　金：国家自然科学基金资助项目(61202298;61300138);福建省自然科学基金资助项目(2014J01239;2015J01656);华侨大学高层次人才科研启动项目(14BS207);华侨大学中青年科研提升计划(ZQNPY309);华侨大学研究生科研创新能力培育计划资助项目(1400414009)

摘　　要：针对原始运动捕获数据结构复杂、语义模糊的问题,提出一种结合金字塔模型和随机森林的运动捕获序列语义标注方法.首先,利用概率主成分分析将运动序列划分为具有特定语义的运动片段.然后,将运动片段的欧拉角数据转换为人体各个关节点的三维空间位置坐标数据,统一骨骼长度,提取运动数据的2种互补性几何特征,并分别归一化.再次,运用傅里叶时间金字塔模型构建运动片段完整的时空特征.最后,利用已训练的随机森林分类器对各个运动片段进行标注.结果表明:该方法能够对具有不同语义的复杂运动序列进行有效标注,且可用于不同表演者,具有一定的实用性和通用性.According to the complexity and semantic ambiguity within original motion capture data,we presents an effective motion capture sequence semantic approach via pyramid model and random forests.Firstly,utilize probabilistic principal component analysis to segment motion sequences into several motion clips with certain semantics.Then,the Euler angler data of each motion clip are transformed into three-dimensional space coordinates of each human joint,and the bone lengths are unified.Subsequently,two complementary features are extracted and normalized.Accordingly,the Fourier temporal pyramid model is adopted to represent the spatiotemporal characteristics of motion clips.Finally,the trained random forests classifier is employed to label each motion clip.The proposed approach is able to well annotate complex motion sequences effectively and can be applied to different performers.The experimental results show that it has certain practically and generality.

关 键 词：语义标注 概率主成分分析 傅里叶时间金字塔 随机森林 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]