基于改进深度学习的人体姿态识别方法研究
    点此下载全文
引用本文:刘宇.基于改进深度学习的人体姿态识别方法研究[J].计算技术与自动化,2024,(2):182-186
摘要点击次数: 28
全文下载次数: 0
作者单位
刘宇 (清华大学 信息科学技术学院北京 100062) 
中文摘要:针对现有2D多人人体姿态识别方法存在的耗时长、准确率低等问题,在对人体姿态识别方法进行分析的基础上,提出了一种用于2D多人人体姿态识别的改进复合场。引入空洞卷积模块降低参数量的同时提高模型准确性,引入shuffleNet V2网络替换主干网ResNet提高模型识别速度。通过实验对所提方法的平均精确度、平均召回率和运行时间等进行分析。结果表明,与常规方法相比,所提方法对2D多人人体姿态识别具有较高的识别准确率和速度,1~8人的人体姿态平均识别时间为75ms。为计算机视觉的研究提供了一定的参考。
中文关键词:人体姿态  复合场  空洞卷积模块  shuffleNet V2 网络  2D多人
 
Research on Human Posture Recognition Method Based on Improved Deep Learning
Abstract:Based on the analysis of existing 2D multi human body pose recognition methods, an improved composite field for 2D multi human body pose recognition is proposed to address the issues of long time consumption and low accuracy.Introducing a hollow convolutional module to reduce the number of parameters while improving model accuracy, and introducing shuffleNet V2 network to replace the backbone network ResNet to improve model recognition speed.Analyze the average accuracy, average recall rate, and running time of the proposed method through experiments.The results show that compared with conventional methods, the proposed method has higher recognition accuracy and speed for 2D multi person human pose recognition, with an average recognition time of 75ms for 1-8 people. This provides a certain reference for the research of computer vision.
keywords:body posture  composite field  atrous convolution module  ShuffleNet V2 network  2D multiple persons
查看全文   查看/发表评论   下载pdf阅读器