摘要
随着现在人体的运动捕获和行为理解的研究的发展,对这项研究有了越来越高的要求。相对于原来的手动提取人体关节点作为特征点来研究,如何使得提取特征点更加自动化,对以后的运动捕获和行为理解的研究意义重大。提出一种在单目视觉条件下在第一帧自动提取人体关节点位置的方法,来解决传统的以手动标定提取人体关节点的问题,并且利用光流稀疏L_K算法对提取出的关节点进行运动跟踪,得到运动人体二维坐标信息,结合像机模型通过几何计算获得人体关节点的深度信息。
Now with the development of the research of the motion capture and the behavior understanding of the human,the demand is increasing.The people extracted the articulation points as the feature points manually before,compared with that.How to make the extraction of the feature points much more automated is meaningful to the research of the motion capture and the behavior understanding later.This paper proposed the method of extracting the dynamic human's articulation points and obtaining the position of the articulation points on the first frame image automatically from monocular video sequences to solve the disadvantage of traditional method about extracting the points from marked human body manually.Then the points were tracked with the light flow sparse L_K algorithm to acquire the information of two-dimensional coordinate about the moving human.Finally,combining the camera model the relative depth of the points can be obtained through geometric calculation.
出处
《计算机科学》
CSCD
北大核心
2013年第4期292-294,共3页
Computer Science
关键词
关节点
比例正交投影
比例因子
重建
Articulation points
Scaled-orthographic camera model
Scale factor
Reconstruction