基于多源信息融合的作业动作识别技术

打开文本图片集

关键词：动作识别；深度学习；信息融合；图像处理；惯性测量

DOI：10. 15938/j. jhust. 2025.06.005

中图分类号：TP751 文献标志码：A 文章编号：1007-2683（2025）06-0043-08

Abstract：Withthedevelopmentof deep learningandcomputerhardware，theapplicationof virtualrealityfortrainingand assessentof practitioners isamajortrend.Through efective human motionrecognition，operatorscan practicecomplexoperation processes invalevironents.However，heurentmehodoftaskactioecoitionstillelsoaiglesnsor，ichishighly susceptible tointerferenceandcanleadtomisjudgmentandmisjudgement.Thisarticlefocusesontheaboveisuesandcoducts researchonliveworkactionrecognitiontechnologybasedonmulti-sourceinformationfusionWehavedesignedanIMUsignalfeature extractinmodelbasedon Transformerandavideoactionfeature extraction modelbasedon 3DCNNdeeplearning networkframework， achievingefectivefusionofIUsignalandvideoinformation.Experimentalverificationhasbenconductedinfourtypicaloerating environments，includingtrainingcenters，substations，outdoorareas，andconfinedspaces.Theexperimentproves thatthemethod proposed in this article can efectively identify live working actions.

Keywords ：action recognition； deep learning； information fusion；image processing； inertial measurement

0 引言

电力系统的运行维护操作具有特殊性，对操作正确性的要求非常高，一般不允许出现重大的操作失误，对于复杂的操作流程和高难度的操作技能，需要通过反复的操作训练来提高从业人员的技能水平。（剩余10378字）