基于树形图结构模型的人体姿态估计

Human pose estimation based on tree-like picture model

讨论计算机视觉领域的人体姿态估计问题,基于树形图结构模型提出一种人体姿态估计算法。根据人体目标前景分割图的形状特征给出人体部位外观模型后,利用高斯分布确定人体部位间的结构先验分布,再采用置信传播算法进行人体姿态估计。仿真结果表明新算法有效。

The question of human pose estimation in the domain of computer vision was discussed in this paper. A human pose estimation method was proposed based on tree-like picture struc- ture. A human part appearance model was proposed according to the shape feature of foreground segmentation image. The structure prior distribution was determined by Gauss function and then the belief propagation algorithm was used for estimating the human pose. The simulation results showed the proposed human pose estimation method was effective.