Abstract: The Support Vector Data Description（SVDD） algorithm is one of the best way to solve the problem of single class classification, it has been applied sucessfully in the human pose estimation problem, and good results have been achieved in the establishment of the part appearance model. However, all of the training samples and different feature of sample in the existing part appearance models, which are built by the SVDD algorithm, are treated equally. For overcoming these two defects, a sample and feature weighted SVDD algorithm is proposed, and a part appearance model based on sample and feature weighted SVDD algorithm is constructed. The weighting coefficient of sample is determined by the distance between the sample and sample center, the weighting coefficient of feature is computed according to the number that the corresponding image regions in the training images of sample feature are contained by the real human part. The experiment results show that the proposed part appearance model can represent the appearance of real human part more accurately than the part appearance model based on the standard SVDD algorithm, and the higher accuracy of human pose estimation can be got.

Key words: human pose estimation, part appearance model, Support Vector Data Description（SVDD）, histogram of oriented gradient, weighting coefficient

摘要： 支持向量数据描述（SVDD）算法是解决单类分类问题的最好方法之一，在人体姿态估计问题中获得了成功的应用，在建立部位外观模型方面取得了良好的效果，但现有利用SVDD算法建立的部位外观模型将所有训练样本和样本不同特征都平等对待。为克服存在的这两个缺陷，提出了一种样本和特征加权的SVDD算法，并用其建立了一种基于样本和特征加权SVDD算法的部位外观模型。样本的权重系数根据样本到样本中心的距离远近来确定，样本特征的权重系数根据特征对应图像区域被训练图像中真实人体部位包含次数的多少来确定。仿真实验结果表明所建立的部位外观模型比利用标准SVDD算法建立的部位外观模型能更准确地描述真实人体部位的外观，能得到更高的人体姿态估计准确度。

关键词: 人体姿态估计, 部位外观模型, 支持向量数据描述, 梯度方向直方图, 权重系数