Computer Engineering and Applications ›› 2013, Vol. 49 ›› Issue (14): 30-33.

Previous Articles     Next Articles

Human pose estimation algorithm based on pictorial structure model

HAN Guijin1,2, ZHU Hong1   

  1. 1.Faculty of Automation and Information Engineering, Xi’an University of Technology, Xi’an 710048, China
    2.School of Automation, Xi’an University of Posts and Telecommunications, Xi’an 710121, China
  • Online:2013-07-15 Published:2013-07-31

一种基于图结构模型的人体姿态估计算法

韩贵金1,2,朱  虹1   

  1. 1.西安理工大学 自动化与信息工程学院,西安 710048
    2.西安邮电大学 自动化学院,西安 710121

Abstract: Aiming at the problem of estimating human pose in single image, an efficient algorithm for estimating human pose is proposed, which is based on pictorial structure model. In this paper, a new part appearance model and a new approach for reducing state space are proposed:(1) HOG feature is computed by using cells with different sizes for different parts of the body and linear SVM is used to classify, then a new part appearance model is proposed; (2) location region of parts is determined according to location prior, and then state space is reduced further through merging neighborhood and setting matching degree threshold, thus a new approach for reducing state space is proposed. Compared with additional human pose estimation algorithms, the simulation results show that the algorithm is more efficient.

Key words: human pose estimation, pictorial structure model, Histogram of Oriented Gradient(HOG) feature, state space

摘要: 针对单幅图片中人体姿态的估计问题,在图结构模型的基础上提出了一种新的人体姿态估计算法。算法提出了一个新的部位观测模型和一种新的减小部位状态空间的方法:(1)对人体不同部位采用不同尺寸的细胞单元计算HOG特征,并利用线性SVM进行分类,从而提出一种新的部位观测模型;(2)利用人体部位定位的先验分布确定部位定位区域,然后通过邻域归并和设置与部位模板的匹配度阈值进一步减小状态空间,从而提出了一种减小部位状态空间的方法。仿真实验结果表明所提算法与传统算法相比更加有效。

关键词: 人体姿态估计, 图结构模型, 方向梯度直方图(HOG)特征, 状态空间