三维人体点云模型多约束肢体分割  

Multi-constrained segmentation of 3D human point-cloud

作　　者：张翔宇[1,2] 田庆国[1,2] 葛宝臻[1,2] 

机构地区：[1]天津大学精密仪器与光电子工程学院,天津300072 [2]光电信息技术教育部重点实验室(天津大学),天津300072

出　　处：《计算机应用》2015年第3期830-834,共5页journal of Computer Applications

基　　金：国家自然科学基金面上项目(61177002)

摘　　要：针对人体点云模型的肢体分割这一动作识别和虚拟重建领域的重要问题,提出了一种基于分类骨架线、测地距离、特征点和姿态分析的多约束肢体分割算法,通过生成点云模型的分类骨架线,配合测地距离获得人体各部位粗分割点云集,利用测地路径方法实现关键特征点的定位,并利用曲线拟合方式进行定位优化,针对头颈、上肢、下肢和躯干之间关联部位的解剖学特征,构造多种约束条件,对各部位粗分割点云集进行了优化再分割。实验结果表明,所提算法对站姿条件下的不同动作、不同体型、不同精度人体点云模型均能取得与视觉理解相吻合的分割效果。通过该算法得到的肢体各部分点云数据可用于姿态分析等后续研究。Body parts segmentation of human point-cloud model is an important research content in action recognition and virtual reconstruction fields. Focused on this issue, a multi-constrained segmentation algorithm based on classified skeleton, geodesic distance, feature points and posture analysis was proposed. By generating the classified skeleton and geodesic distance of point-cloud, the roughly segmented point sets of each body part were got. Feature points were positioned by an algorithm depending on geodesic path and optimized by a curve fit method. According to these feature points and some anatomical features of human body, multiple constraints were constructed and roughly segmented point sets were segmented once again. The experimental results demonstrate that the segmentation effects of human point cloud models with different action, size and precision in standing posture are consistent with visual understanding of human. The point-cloud of body parts obtained through this algorithm can be used for posture analysis and so on.

关 键 词：人体点云模型 点云分割 测地距离 姿态分析 多约束 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]