基于高分辨率网络的人体姿态估计
Human pose estimation algorithm based on high-resolution net
投稿时间:2021-12-20  修订日期:2021-12-20
DOI:
中文关键词: 姿态估计  高分辨率网络  空洞卷积  人体检测  关键点相似度
英文关键词: Pose estimation  High-resolution net  Dilated convolution  Human detection  Key point similarity
基金项目:国家自然科学基金资助项目(No.61671483);湖北省自然科学基金资助项目(NO.2016CFA089)
作者单位邮编
朱翠涛 中南民族大学智能无线通信湖北省重点实验室 
李博 中南民族大学智能无线通信湖北省重点实验室 430074
摘要点击次数: 180
全文下载次数: 0
中文摘要:
      针对高分辨率网络中存在不同分支特征交叉融合导致参数量大、运算复杂度高等问题,提出了一种基于高分辨率检测网络(HRNet)的人体姿态估计优化网络模型。首先引入空洞空间卷积池化金字塔替代多分辨率分支网络交叉融合过程,同时引入注意力机制,提高网络输出特征图质量,从而保证改进后网络检测的准确度。在环境配置和网络输入图像分辨率一致的情况下,本模型在COCO数据集上实验结果较HRNet相比参数量下降38.6%,运算复杂度下降35.2%。实验结果表明,改进后网络在检测精度略微下降的情况下,能有效降低参数量、运算复杂度。
英文摘要:
      Aiming at the problems of large amount of parameters and high computational complexity caused by the cross fusion of different branch features in high-resolution network, a human posture estimation optimization network model based on high-resolution detection network (HNRet) is proposed. Firstly, the hole space convolution pool pyramid is introduced to replace the cross fusion process of multi-resolution branch networks, and the attention mechanism is introduced to improve the quality of network output feature map, so as to ensure the accuracy of improved network detection. When the environment configuration is consistent with the resolution of the network input image, the experimental results of the model on the coco data set are 38.6% lower than that of HNRet, and the computational complexity is 35.2%. The experimental results show that the improved network can effectively reduce the amount of parameters and computational complexity when the detection accuracy decreases slightly.
View Fulltext   查看/发表评论  下载PDF阅读器
关闭