融合语义与几何特征的人体模型结构分割  

Structural segmentation of human models based on semantic and geometric characteristics

作　　者：黎琳[1] 韩丽[1,2] 徐建国[1] 唐棣[1] 

机构地区：[1]辽宁师范大学计算机与信息技术学院,辽宁大连116081 [2]大连理工大学计算机科学与技术系,辽宁大连116024

出　　处：《计算机应用研究》2016年第1期316-320,共5页Application Research of Computers

基　　金：辽宁省高等学校优秀人才支持项目(LJQ2013110)

摘　　要：针对现有三维人体模型形状分析方法存在人工干预及对姿势依赖的问题,提出一种融合语义与几何特征的三维人体形状分析方法。基于模型表面测地线距离以及内部空间体积特征的度量,提出了基于骨架树的结构检测方法,基于人体测量学先验语义知识,进一步提炼模型的层次结构。该方法能有效地提取不同姿态人体模型的结构特征,并实现基于语义的模型分割。一系列实验结果验证了方法的高效性与鲁棒性,对于实现模型的高层次语义描述以及模型检索具有重要的意义。Aiming at the problems that current methods of shape analysis for 3 D human models need manual intervention and are not pose-independence, this paper proposed an improved algorithm by combining the semantic knowledge with geometric features. Firstly, the authors adopted geodesic distance and space volume as surface metric and volume metric to provide multi- scale structure detection with a skeleton tree descriptor, then further extracted the hierarchy structure of the human models based on the prior knowledge of anthropometry. The method could extract the structural features of different human posture model effectively, and realized the segmentation model based on semantic. A series of experimental results verifies the efficiency and robustness of this method, and our work has important significance for high level semantic shape descriptor and shape re- trieval.

关 键 词：形状分析 人体模型 测地线距离 层次划分 空间体积 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]