• 首页
  • 期刊简介
  • 编委会
  • 投稿指南
  • 收录情况
  • 杂志订阅
  • 联系我们
引用本文:徐寅哲,屠佳佳,李 洲,史伟民.基于关键点检测的坐姿识别方法[J].软件工程,2024,27(4):33-37.【点击复制】
【打印本页】   【下载PDF全文】   【查看/发表评论】  【下载PDF阅读器】  
←前一篇|后一篇→ 过刊浏览
分享到: 微信 更多
基于关键点检测的坐姿识别方法
徐寅哲, 屠佳佳, 李 洲, 史伟民
(浙江理工大学机械工程学院, 浙江 杭州 310018)
brgzz@qq.com; tjji23@163.com; 2521170001@qq.com; swm@zstu.edu.cn
摘 要: 针对直接应用深度学习分类算法所得模型泛用性较差的问题,提出了一种分两个步骤完成人体坐姿识别的方法。所提方法首先提取图片中人体上身关键点的坐标信息,在关键点检测环节采用高分辨率主干网络,并进一步改进了模型结构,在下采样环节中引入SE(Squeeze-and-Excitation)注意力机制,加强了空间位置特征的表达,取得了更高的检测平均精准度;然后采用随机森林算法对关键点进行坐姿分类。实验结果表明:所提方法识别准确率可以达到94%以上,并且在陌生场景下有更好的泛用性,能适应实际应用中复杂的人物环境。
关键词: 坐姿识别;高分辨率网络;人体关键点检测;随机森林算法
中图分类号: TP391.41    文献标识码: A
Sitting Posture Recognition Based on Keypoints Detection
XU Yinzhe, TU Jiajia, LI Zhou, SHI Weimin
(College of Mechanical Engineering, Zhejiang Sci-Tech University, Hangzhou 310018, China)
brgzz@qq.com; tjji23@163.com; 2521170001@qq.com; swm@zstu.edu.cn
Abstract: Aiming at the problem of poor generalizability of the model obtained from the direct application of deep learning classification algorithm, this paper proposes a two-step method to complete the human sitting posture recognition. In the proposed method, firstly the coordinate information of the keypoints of the upper human body in the picture is extracted, a high resolution backbone network in the keypoints detection link is adopted, and the model structure is furtherly improved. The Squeeze-and-Excitation (SE) attention mechanism is introduced in the downsampling link to strengthen the expression of spatial positional features, achieving a higher detection average accuracy. Then the Random Forest algorithm is used to classify the keypoints for sitting posture. Experimental results show that the recognition accuracy of the proposed model can reach over 94% , and it has better generalization in unfamiliar scenarios, which better adapts to environments with complex characters in practical applications.
Keywords: sitting posture recognition; high resolution network; human keypoints detection; Random Forest algorithm


版权所有:软件工程杂志社
地址:辽宁省沈阳市浑南区新秀街2号 邮政编码:110179
电话:0411-84767887 传真:0411-84835089 Email:semagazine@neusoft.edu.cn
备案号:辽ICP备17007376号-1
技术支持:北京勤云科技发展有限公司

用微信扫一扫

用微信扫一扫