摘要
采用自顶而下的方法将整个框架分解为两个分支。第一分支为目标识别器,输出人体语义特征图;第二分支生成姿势热图,通过增加特征筛选因素来提高识别精度。通过融合算法,联合两个分支的置信度并构造损失函数,进而获取较高精度的人体姿势估计图。
A top-down method is proposed for human position estimation.The framework is divided into two branches,in which the first one is for target reading to output the semantic characteristics graph of human body,while the other generates the pose heat maps to improve identification accuracy by adding feature screening.With the fusion algorithm,the confidence of the two branches are combined to construct a loss function for getting the high precision human pose estimation.
作者
郭昕刚
梁锦明
GUO Xingang;LIANG Jinming(School of Computer Science & Engineering, Changchun University of Technology, Changchun 130012, China)
出处
《长春工业大学学报》
CAS
2020年第6期591-596,共6页
Journal of Changchun University of Technology
基金
吉林省发改委项目(2019C040-3)。
关键词
人体姿势估计
人体目标分割
语义特征图
损失函数
human pose estimation
object segmentation
semantic feature map
loss function
