文献类型：期刊文献

中文题名：基于高阶空间交互作用的姿态估计网络

英文题名：Pose Estimation Network Based on High-Order Spatial Interactions

作者：黄晓宇[1];陈佳艺[1];吴艺玮[1];吴胜昔[1];王学武[1]

机构：[1]华东理工大学能源化工过程智能制造教育部重点实验室,上海200237

年份：2025

卷号：51

期号：4

起止页码：514

中文期刊名：华东理工大学学报(自然科学版)

外文期刊名：Journal of East China University of Science and Technology

收录：;北大核心:【北大核心2023】；

基金：国家自然科学基金(62076095)。

语种：中文

中文关键词：姿态估计;高分辨率网络;高阶空间交互;CBAM注意力机制;特征提取

外文关键词：pose estimation;high-resolution network;high-order spatial interaction;CBAM attention mechanism;feature extraction

摘要：人体姿态估计是计算机视觉领域的一个重要研究方向。随着深度学习技术的进步,现有的姿态估计模型在预测人体关键点方面已经取得了显著成效,然而,在处理复杂场景如严重遮挡、复杂背景、极端姿态、多尺度变化和光照变化时,这些模型仍然面临挑战,准确度往往受到影响。为解决这个问题,本文提出了一种改进的基于高分辨率网络(High-Resolution Network,HRNet)的人体姿态估计方法,该方法通过引入高阶空间交互和注意力机制,显著提升了模型在复杂场景中的表现;并采用递归门控卷积和卷积注意力模块以增强模型在高阶空间特征提取的能力。结果表明,提出的方法在COCO2017数据集上超越了现有主流方法,实现了更高的姿态估计精度。
Human pose estimation is a crucial research area in computer vision.With the advancement of deep learning technologies,existing pose estimation models have achieved remarkable success in predicting human keypoints.However,when dealing with complex scenes such as severe occlusion,complex backgrounds,extreme poses,multi-scale variations,and lighting changes,these models still face challenges and their accuracy is often affected.To address this issue,this paper proposes an improved human pose estimation method based on HRNet,which significantly improves the performance of the model in complex scenes by introducing high-order spatial interaction and attention mechanisms.It employs recursive gated convolution and convolutional attention modules to enhance the model's ability to extract high-order spatial features.The experimental results show that the proposed method outperforms existing mainstream approaches on the COCO2017 dataset and achieves higher pose estimation accuracy.