基于显著性特征的多视角动作图像识别研究

打开文本图片集

引用格式：.基于显著性特征的多视角动作图像识别研究[J].现代电子技术，2025，48（13）：62-65.

Research on multi view action image recognition based on saliency features

HUIXianghui1，SUNYanhong1，SHENXiaole² （1.Collegeof Informationand ManagementScience，Henan Agricultural UniversityZhengzhou 45Oo46，China; 2.HenanAgricultural University，Zhengzhou 45oo46，China）

Abstract：Amulti viewactionimagerecognitionmethodbasedonsaliencyfeaturesisstudied.Thekeyfeaturesof athlete actionsareautomaticallylearnedandextracted，whichcanhelpcoaches develop morescientificandpersonalized trainingplans forathletes.Thehuman skeletonsequenceisaligned toaunified spatiotemporal coordinate system，and both thedistance map andangle maparecalculated to capture thespatialfeaturesoftheskeleton，soastogenerateahuman motion feature map.A CNN + CA modelisbuilt.Aregionof interest （ROI）splicingdiagram fromtheprocessedmultiviewactionvideo framesis generatedandinputintoCNN，andthenmultiviewfusionfeaturesareextracted.Inaddition，themostcriticalregionforaction imagerecognitionintheCAmoduleishighlighted.Bysequencematchingalgorithm，themultiviewactionrecognitionis transformedintoamatchingofpredictinglabelsequences，andactioncategorylabelsareassignedtotheactionimagestobe recognized，whichachievesaccuratemultiviewactionimagerecognition.Theexperimentalresultsshowthatthemethodcannot onlyeffectivelyproesactioimagesinerentperspeties，utalsoccuratelyecogeuliletiosofsketballplas.

Keywords：salient feature; multi view action image; motion feature map;ROI splicing diagram; CNN; CA module; LSTM; sequence matching algorithm

0 引言

在体育领域中，准确捕捉和分析运动员的动作是提升训练效果、优化比赛策略以及评估运动员表现的关键环节。（剩余5287字）