计算机技术与发展Issue(1)：111-114,118,5.DOI:10.3969/j.issn.1673-629X.2016.01.023
基于多变量标签分布的连续型姿态估计方法
A Continuous Head Pose Estimation Method Based on Multivariate Label Distribution
摘要
Abstract
The human’s head pose are abundant of information. Accurate estimation of head pose plays an important role in face recogni-tion,expression recognition and so on. To improve the precision of estimation and to alleviate the problem that poses are always fixed to some angles,a continuous method based on multivariate label distribution to estimate head poses was presented. In the training phase,get the discrete multivariate distribution from discrete poses and angles. In the testing phase,adopt orthogonal polynomial fitting to transform the discrete distribution into continuous distribution and compute the label corresponding to maximum in distribution as final output. The proposed method has been tested on the open Pointing’04 database. After orthogonal polynomial fitting,the estimated angles are no lon-ger limited to angles in test set,but more continuous values. The estimated angle obtained by the method proposed is closer to the real an-gle. The result indicates that this method can estimate head pose in wider angle,the result is more stable.关键词
姿态估计/多变量标签分布/正交多项式拟合/连续型姿态Key words
head pose estimation/multivariate label distribution/orthogonal polynomial fitting/continuous head pose分类
信息技术与安全科学引用本文复制引用
江婷,赵启军,陈虎..基于多变量标签分布的连续型姿态估计方法[J].计算机技术与发展,2016,(1):111-114,118,5.基金项目
国家自然科学基金资助项目(61202160,61202161) （61202160,61202161）
科技部重大仪器专项(2013YQ49087904) （2013YQ49087904）
