基于OpenPose与AT-STGCN的电力作业人员行为识别技术  

Behavior Recognition Technology of Power Operator Based on OpenPose and AT-STGCN

作　　者：王鸿 陈明举 熊兴中 张劲松 WANG Hong;CHEN Mingju;XIONG Xingzhong;ZHANG Jinsong(School of Automation and Information Engineering,Sichuan University of Science&Engineering,Yibin 644000,China;Artificial Intelligence Key Laboratory of Sichuan Province,Yibin 644000,China)

机构地区：[1]四川轻化工大学自动化与信息工程学院,四川宜宾644000 [2]人工智能四川省重点实验室,四川宜宾644000

出　　处：《四川轻化工大学学报（自然科学版）》2023年第4期61-70,共10页Journal of Sichuan University of Science & Engineering(Natural Science Edition)

基　　金：四川省科技成果转移转化示范项目(2022ZHCG0035);人工智能四川省重点实验室项目(2020RZY02;2021RYY04);企业信息化与物联网测控技术四川省高校重点实验室开放基金项目(2021WYY01);四川轻化工大学研究生创新基金项目(Y2021073)。

摘　　要：为实现复杂环境中作业现场人员的行为的智能识别,提出了一种OpenPose与ATSTGCN结合的作业人员行为识别算法。该网络首先采用OpenPose技术从视频帧中提取人体骨骼关键节点,并采用部分亲和字段连接关键节点获得人体骨骼图,将时空注意力机制引入到时空图卷积网络中构建AT-STGCN模型,实现对人体骨骼图特征提取的能力,以提高人体动作识别精度。实验结果表明,本文构建的模型算法对电力作业人员的动作识别率达到97.70%,相比STGCN提高0.90%,且浮点运算数降低6.45 G,其整体指标优于其他算法,能实现对作业人员行为有效安全监控,具有一定的鲁棒性与泛化能力。In order to realize the intelligent recognition of personnel behavior in complex environment,an operator behavior recognition algorithm combining OpenPose and AT-STGCN has been proposed.Firstly,the key nodes of human bones are extracted from video frames using the OpenPose technology in the network,and the human skeleton map is obtained by connecting the key nodes with some affinity fields.Then,the spatio-temporal attention mechanism is introduced into the spatio-temporal graph convolution network to construct the ATSTGCN model,so as to achieve the ability of extracting the features of human skeleton map and improve the accuracy of human action recognition.The experimental results show that the recognition rate of the electric power operators action using the model algorithm constructed in the present study reaches 97.70%,which is 0.90%higher than that of STGCN,whose floating point arithmetic number is 6.45 G lower than that of STGCN.Moreover,the overall index of the model algorithm is better than that of other algorithms,which can realize effective safety monitoring of the behavior of operators,and exhibits certain robustness and generalization ability.

关 键 词：时空图卷积 注意力机制 人体骨骼图 行为识别 电力作业人员 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]