Abstract: Human model is a mathematical description of the appearance and spatial correlation between the human parts or joints. Although some human models have taken into account the fact that the spatial locations of human parts or joints satisfy a certain prior distribution, the human model based on the same prior distribution is used to different images for human pose estimation, and does not take into account the location difference of different images. For overcoming this deficiency, a human model using conditional prior is proposed, an adaptive adjustment parameter is added to the human model. The size of the parameter is adjusted adaptively according to the size of the similarity between the potential location and the appearance model, thus when calculating the joint localization probability, the purpose of adjusting the prior distribution according to the picture to be processed is achieved. The human model based on conditional prior is used for human pose estimation, the simulation experiments show that the proposed human model can obtain higher estimation accuracy than the existing human model based on prior.

Key words: human pose estimation, human model, conditional prior, adaptive adjustment parameter

摘要： 人体姿态估计算法中的人体模型是对人体部位或关节间外观和空间关联情况的数学描述。虽然当前已经有部分人体模型在建立时考虑到了部位或关节的空间定位会满足一定的先验分布，但却都将基于同样先验分布建立的人体模型用于不同待处理图片进行人体姿态估计，并没有考虑到不同图片中的实际定位对先验分布服从程度的不同。为此，提出了一种基于条件先验的人体模型，在人体模型中添加自适应调节参数，参数值根据待处理图片中关节可能定位与外观模型相似度的大小来确定，从而达到根据待处理图片自适应调节先验分布在计算关节定位概率时所起作用大小的目的。将基于条件先验的人体模型用于人体姿态估计，仿真实验表明，与现有基于先验的人体模型相比，所提人体模型用于人体姿态估计时获得了更高的估计准确度。

关键词: 人体姿态估计, 人体模型, 条件先验, 自适应调节参数