基于昇腾处理器的边端人体动作识别算法设计与实现

TP391; 针对现有的人体动作识别算法精度不足、计算量大、缺少在边端设备上的部署等问题,本文提出一种基于昇腾处理器的边端轻量化人体动作识别时空图卷积算法.通过设计隐性联系骨架连接方法并构建隐性邻接矩阵,结合自然骨架连接邻接矩阵,构造显隐性融合空间图卷积.在时间维度加入空间注意力机制,使模型关注不同帧间关节点位置空间特征,进一步设计时间图卷积,构建时空图卷积.此外设计网络中的Ascend-Enisum算子,进行张量融合运算,降低了计算复杂度,使模型轻量化.针对上述改进,在KTH数据集上进行实验验证,与经典单流算法ST-GCN相比,模型计算量减小了22.28%,Top-1精度达到84.17%,提...

Full description

Saved in:
Bibliographic Details
Published in光电工程 Vol. 51; no. 6; pp. 中插5 - 72
Main Authors 赵冬冬, 赖亮, 陈朋, 周鸿超, 李亦然, 梁荣华
Format Journal Article
LanguageChinese
Published 浙江工业大学计算机科学与技术学院,浙江 杭州 310023 01.06.2024
Subjects
边端人体动作识别
昇腾处理器
edge human action recognition
ascend processor
spatial temporal graph convolutional
时空图卷积
轻量化
lightweight
Online AccessGet full text
ISSN1003-501X
DOI10.12086/oee.2024.240072

Cover

Abstract TP391; 针对现有的人体动作识别算法精度不足、计算量大、缺少在边端设备上的部署等问题,本文提出一种基于昇腾处理器的边端轻量化人体动作识别时空图卷积算法.通过设计隐性联系骨架连接方法并构建隐性邻接矩阵,结合自然骨架连接邻接矩阵,构造显隐性融合空间图卷积.在时间维度加入空间注意力机制,使模型关注不同帧间关节点位置空间特征,进一步设计时间图卷积,构建时空图卷积.此外设计网络中的Ascend-Enisum算子,进行张量融合运算,降低了计算复杂度,使模型轻量化.针对上述改进,在KTH数据集上进行实验验证,与经典单流算法ST-GCN相比,模型计算量减小了22.28%,Top-1精度达到84.17%,提升了5% .基于上述算法设计了昇腾AI人体动作识别系统,并在边端设备成功部署,可以进行实时人体动作识别.
AbstractList TP391; 针对现有的人体动作识别算法精度不足、计算量大、缺少在边端设备上的部署等问题,本文提出一种基于昇腾处理器的边端轻量化人体动作识别时空图卷积算法.通过设计隐性联系骨架连接方法并构建隐性邻接矩阵,结合自然骨架连接邻接矩阵,构造显隐性融合空间图卷积.在时间维度加入空间注意力机制,使模型关注不同帧间关节点位置空间特征,进一步设计时间图卷积,构建时空图卷积.此外设计网络中的Ascend-Enisum算子,进行张量融合运算,降低了计算复杂度,使模型轻量化.针对上述改进,在KTH数据集上进行实验验证,与经典单流算法ST-GCN相比,模型计算量减小了22.28%,Top-1精度达到84.17%,提升了5% .基于上述算法设计了昇腾AI人体动作识别系统,并在边端设备成功部署,可以进行实时人体动作识别.
Abstract_FL Aiming at the problems of existing human action recognition algorithms such as insufficient accuracy,large amount of calculation,and lack of deployment on edge devices,this paper proposes an edge-side lightweight human action recognition spatial temporal graph convolutional algorithm based on the Ascend processor.By designing an implicit skeletal connection method and constructing an implicit adjacency matrix,combined with the natural skeletal connection adjacency matrix,we create an explicit-implicit fusion spatial graph convolution.A spatial attention mechanism is added to the temporal dimension,enabling the model to focus on spatial features of joint positions across different frames.Furthermore,we design a temporal graph convolution to construct a spatiotemporal graph convolution.Additionally,the Ascend-Enisum operator is designed within the network to perform tensor fusion operations,reducing computational complexity and lightening the model.Experimental validation on the KTH dataset demonstrates that,compared to the classical single-stream ST-GCN algorithm,our model achieves a 22.28% reduction in computational cost while attaining a Top-1 accuracy of 84.17%,representing a 5% improvement.Based on this algorithm,we have designed the Ascend AI human action recognition system,which has been successfully deployed on edge devices for real-time human action recognition.
Author 赵冬冬
陈朋
赖亮
李亦然
梁荣华
周鸿超
AuthorAffiliation 浙江工业大学计算机科学与技术学院,浙江 杭州 310023
AuthorAffiliation_xml – name: 浙江工业大学计算机科学与技术学院,浙江 杭州 310023
Author_FL Zhao Dongdong
Zhou Hongchao
Liang Ronghua
Lai Liang
Chen Peng
Li Yiran
Author_FL_xml – sequence: 1
  fullname: Zhao Dongdong
– sequence: 2
  fullname: Lai Liang
– sequence: 3
  fullname: Chen Peng
– sequence: 4
  fullname: Zhou Hongchao
– sequence: 5
  fullname: Li Yiran
– sequence: 6
  fullname: Liang Ronghua
Author_xml – sequence: 1
  fullname: 赵冬冬
– sequence: 2
  fullname: 赖亮
– sequence: 3
  fullname: 陈朋
– sequence: 4
  fullname: 周鸿超
– sequence: 5
  fullname: 李亦然
– sequence: 6
  fullname: 梁荣华
BookMark eNotj8tKw0AYhWdRwVq79wVcJv4zmUwmSyn1AgU3Cu7KpDMTFEnAIL5AreKiCCJeEEVwUSikghsNNE-TafIYRnR1Fh-c75wV1IjiSCG0hsHGBDjbiJWyCRBqEwrgkQZqYgDHcgEfLqN2khwFAJT4HmdeE3XNa1Zk48XDZXWRm_dheTMyj5PyaVjl3-V0VmRZMb8115Ni_lzNRuZqWqb3i8-7Ks2r9K34Gpv0pRx_rKIlLU4S1f7PFjrY6u53dqze3vZuZ7NnJfUEZjke00AV1cCloEQHoDmTAxcTqonEvq4ZI0xIIWrkySDwfcBEceK5PlfYaaH1v95zEWkRhf3j-Ow0qo39UIaD39PAapHzA0F9ZvA
ClassificationCodes TP391
ContentType Journal Article
Copyright Copyright © Wanfang Data Co. Ltd. All Rights Reserved.
Copyright_xml – notice: Copyright © Wanfang Data Co. Ltd. All Rights Reserved.
DBID 2B.
4A8
92I
93N
PSX
TCJ
DOI 10.12086/oee.2024.240072
DatabaseName Wanfang Data Journals - Hong Kong
WANFANG Data Centre
Wanfang Data Journals
万方数据期刊 - 香港版
China Online Journals (COJ)
China Online Journals (COJ)
DatabaseTitleList
DeliveryMethod fulltext_linktorsrc
DocumentTitle_FL Design and implementation of edge-based human action recognition algorithm based on ascend processor
EndPage 72
ExternalDocumentID gdgc202406006
GroupedDBID 2B.
4A8
8FE
8FG
92I
93N
ABJCF
ABUWG
AFKRA
ALMA_UNASSIGNED_HOLDINGS
BENPR
BGLVJ
BPHCQ
BVBZV
CCPQU
HCIFZ
L6V
M7S
PHGZM
PHGZT
PIMPY
PMFND
PQQKQ
PROAC
PSX
PTHSS
TCJ
ID FETCH-LOGICAL-s1006-376f04e4f08da42fb0f86dc5124f2d19f4e4626adaab0f7dbb99012e827598e13
ISSN 1003-501X
IngestDate Thu May 29 03:55:49 EDT 2025
IsPeerReviewed true
IsScholarly true
Issue 6
Keywords 边端人体动作识别
昇腾处理器
edge human action recognition
ascend processor
spatial temporal graph convolutional
时空图卷积
轻量化
lightweight
Language Chinese
LinkModel OpenURL
MergedId FETCHMERGED-LOGICAL-s1006-376f04e4f08da42fb0f86dc5124f2d19f4e4626adaab0f7dbb99012e827598e13
ParticipantIDs wanfang_journals_gdgc202406006
PublicationCentury 2000
PublicationDate 2024-06-01
PublicationDateYYYYMMDD 2024-06-01
PublicationDate_xml – month: 06
  year: 2024
  text: 2024-06-01
  day: 01
PublicationDecade 2020
PublicationTitle 光电工程
PublicationTitle_FL Opto-Electronic Engineering
PublicationYear 2024
Publisher 浙江工业大学计算机科学与技术学院,浙江 杭州 310023
Publisher_xml – name: 浙江工业大学计算机科学与技术学院,浙江 杭州 310023
SSID ssib004297867
ssib023646518
ssib036437391
ssib023167165
ssj0002964646
ssib002258422
ssib001102639
ssib051369860
ssib000459782
Score 2.4295084
Snippet TP391; 针对现有的人体动作识别算法精度不足、计算量大、缺少在边端设备上的部署等问题,本文提出一种基于昇腾处理器的边端轻量化人体动作识别时空图卷积算法.通过设计隐性联...
SourceID wanfang
SourceType Aggregation Database
StartPage 中插5
Title 基于昇腾处理器的边端人体动作识别算法设计与实现
URI https://d.wanfangdata.com.cn/periodical/gdgc202406006
Volume 51
hasFullText 1
inHoldings 1
isFullTextHit
isPrint
journalDatabaseRights – providerCode: PRVPQU
  databaseName: ProQuest Technology Collection
  issn: 1003-501X
  databaseCode: 8FG
  dateStart: 20180101
  customDbUrl:
  isFulltext: true
  dateEnd: 99991231
  titleUrlDefault: https://search.proquest.com/technologycollection1
  omitProxy: true
  ssIdentifier: ssj0002964646
  providerName: ProQuest
link http://utb.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwnR1Na9RANNR68SKKip-lB-dUtiaTyWTmONkmFkFPLfRWNk1SvWzBbi891yoeiiDiB6IIHgqFreBFF9pfs9vdn-F7L9NN2gp-wDLMzrz39n1MMm925r1xnLthIWBk5H4jy2AECyF0QxUiaOCmU8sXKs04RiM_fCTnF8WDpWBp4lw9Q_BGJ51d2fxtXMn_WBXawK4YJfsPlh0ThQaog32hBAtD-Vc2ZnHAdMIiw2KBpYpZLJlWDFb4MZQBi2KEMeAwChaHTLtMScLSzChqMdSlEDLS2GIiZhJL0FKeY9pHLGUIq2xpIhZAlgSVQkREj5kOkY3IZzogmJjYoIrxCF0RqwEBx4gFXyO37igTTfgQSxpQAmyJgH5AvwIUxptFxH5gQYAd0xxXToFoaQUzcdWjURvAP6queYIsIHgksUaWo4SoSFbe_nP8XwkX1ZkuGt0kPKkYKx5aqGKdhAell1YxIVXmmJE1BdWUCAyhDUKEBF4q4LESJVpFuRbYJDUYEsxweIrOsjRDCHMIWjKn4xnchSnDsu00hScKA5cuGxrPYzZx75Mzk5LlCQkCTwnTZai59XfKm5POzKTcpZ29tRxzyXIxi2eNS9BT-clXs9UVhHAlpb4_z0Mp8SoRldyvrxt0_doE8Dm5rKWjhUlFiVoeSXCSQlX5xRzTNXjVdjxeeiCDyqnycffZr5I-BZ4vtbLrfnTJ8GyBLCMNjzVnjyygmPdOCUnxe-2i1V6tuZoLl5yLdo04bcoH_rIzsfn4ihMPPvf6vZ2jd89Hzw4HX7eGr7YH73eHH7ZGhz-He_v9Xq9_8Hrwcrd_8HG0vz14sTfsvj36_mbUPRx1v_R_7Ay6n4Y73646i0m80Jxv2DtQGuse_tkH83_hilwUrspaghepWyiZrYCbLgqeebqAPsllK2u1oCvM0hQ3unmueBholXv-NWeyvdbOrzvTYSq9nKfczRWsgVy_VfgF6C2DNRi8rbl7w5myYi_bd9z68gnj3vwTwC3nQvXU3XYmO0838jvgs3fSKRoPvwD0_61e
linkProvider ProQuest
openUrl ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.atitle=%E5%9F%BA%E4%BA%8E%E6%98%87%E8%85%BE%E5%A4%84%E7%90%86%E5%99%A8%E7%9A%84%E8%BE%B9%E7%AB%AF%E4%BA%BA%E4%BD%93%E5%8A%A8%E4%BD%9C%E8%AF%86%E5%88%AB%E7%AE%97%E6%B3%95%E8%AE%BE%E8%AE%A1%E4%B8%8E%E5%AE%9E%E7%8E%B0&rft.jtitle=%E5%85%89%E7%94%B5%E5%B7%A5%E7%A8%8B&rft.au=%E8%B5%B5%E5%86%AC%E5%86%AC&rft.au=%E8%B5%96%E4%BA%AE&rft.au=%E9%99%88%E6%9C%8B&rft.au=%E5%91%A8%E9%B8%BF%E8%B6%85&rft.date=2024-06-01&rft.pub=%E6%B5%99%E6%B1%9F%E5%B7%A5%E4%B8%9A%E5%A4%A7%E5%AD%A6%E8%AE%A1%E7%AE%97%E6%9C%BA%E7%A7%91%E5%AD%A6%E4%B8%8E%E6%8A%80%E6%9C%AF%E5%AD%A6%E9%99%A2%2C%E6%B5%99%E6%B1%9F+%E6%9D%AD%E5%B7%9E+310023&rft.issn=1003-501X&rft.volume=51&rft.issue=6&rft.spage=%E4%B8%AD%E6%8F%925&rft.epage=72&rft_id=info:doi/10.12086%2Foee.2024.240072&rft.externalDocID=gdgc202406006
thumbnail_s http://utb.summon.serialssolutions.com/2.0.0/image/custom?url=http%3A%2F%2Fwww.wanfangdata.com.cn%2Fimages%2FPeriodicalImages%2Fgdgc%2Fgdgc.jpg