基于时域分割的人体行为连续性动作预测仿真  

Prediction and Simulation of Human Behavior Continuity Based on Time Domain Segmentation

作　　者：李丽[1] 庄庆华[1] LI Li;ZHUANG Qing-hua(College of Humanities&Information,Changchun University of Technology,Changchun Jilin 130000,China)

机构地区：[1]长春工业大学人文信息学院,吉林长春130000

出　　处：《计算机仿真》2021年第5期339-343,共5页Computer Simulation

摘　　要：传统人体行为动作预测方法对人体骨架上的所有关键特征映射点包含范围模糊,导致实时处理速度慢,无法有效对连续性动作进行预测,且错位率较高。为此提出一种基于时域分割的人类行为连续性动作预测。首先通过三维空间捕捉技术构建人体骨架模型,通过时域分割确定关节点的位置信息,将任意关节点都视为映射特征点,提取人体的运动数据特征,利用LLE把提取的特征数据映射进二维空间里,结合拉格朗日的乘法最优化重建权值矩阵,建立动作向量库,确定人体做出随意性动作与向量库中动作相似或者相同时,就能够对人体的动作进行提前预判,即可以完成动作预测。仿真结果证明,所提方法时域分割效果好,实时预测速度快,能够有效的对人体连续性动作精准预测。Due to slow real-time processing performance and high malposition rate,this paper put forward a method to predict continuous actions in human behavior based on temporal segmentation.Firstly,the human skeleton model was constructed by three-dimensional space capture technology.The position information of the joint point was determined by temporal segmentation.Any joint points could be regarded as the mapping feature points.Secondly,the motion data features of the human body were extracted.Then,LLE was used to map the extracted feature data into two-dimensional space.Combined with the optimization theory of Lagrangian multiplication,the weight matrix was reconstructed and the action vector library was built.When the random actions were similar or identical to the actions in the vector library,the actions could be predicted in advance.Thus,the motion prediction was completed.Simulation results prove that the proposed method has a good temporal segmentation effect and fast real-time prediction,which can effectively predict the continuous actions of the human body.

关 键 词：时域分割 人体模型提取方法 人体动作识别 动作预测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]