基于Yolov7_Pose的轻量化人体姿态估计网络  

Lightweight human pose estimation network based on Yolov7_Pose

作　　者：黄健[1] 胡翻 展越 HUANG Jian;HU Fan;ZHAN Yue(College of Communication and Information Engineering,Xi’an University of Science and Technology,Xi’an 710600,China)

机构地区：[1]西安科技大学通信与信息工程学院,陕西西安710600

出　　处：《现代电子技术》2024年第23期98-104,共7页Modern Electronics Technique

基　　金：陕西省重点研发计划(2023-YBGY-255);陕西省科技厅工业攻关项目(2022GY-115)。

摘　　要：人体姿态估计在计算机视觉、人机交互与运动分析等领域广泛应用。当前人体姿态估计算法往往通过构建复杂的网络来提高精度,但这带来了模型体量和计算量增大,以及检测速度变慢等问题。因此,文中提出一种基于Yolov7_Pose的轻量化人体姿态估计网络。首先,采用轻量化CARAFE模块替换原网络中的上采样模块,完成上采样工作;接着,在特征融合部分引入轻量化Slim-neck模块,以降低模型的计算量和复杂度;最后,提出了RFB-NAM模块,将其添加到主干网络中,用以获取多个不同尺度的特征信息,扩大感受野,提高特征提取能力。实验结果表明,改进后网络模型的GFLOPs和模型大小分别降低了约18.1%、22%,检测速度提升37.93%,并在低光环境、小目标、密集人群和俯视角度下表现出了较好的性能。Human pose estimation is widely used in computer vision,human⁃computer interaction(HCI)and motion analysis.Current human pose estimation algorithms often improve accuracy by constructing complex networks,but this brings increased model size and computation,as well as slower detection speed.Therefore,this paper proposes a lightweight human pose estimation network based on Yolov7_Pose.A lightweight CARAFE module is used to replace the up⁃sampling module in the original network to complete the up⁃sampling first,and then a lightweight Slim⁃neck module is introduced in the feature fusion section to reduce the computation and complexity of the model.Finally,the RFB⁃NAM module is proposed and added to the backbone network for acquiring feature information at multiple different scales,expanding the receptive field,as well as improving the feature extraction capability.The experimental results show that the computational burden and model size of the improved network model have been reduced by about 18.1%and 22%,respectively,and its detection speed has increased by 37.93%.In addition,it shows better performance in low⁃light environments,detection of small objects,dense crowds,and perspective of overlooking.

关 键 词：人体姿态估计 Yolov7_Pose 轻量化 上采样 CARAFE Slim-neck 

分 类 号：TN911.1-34[电子电信—通信与信息系统] TP391[电子电信—信息与通信工程]