强鲁棒性的可穿戴传感器的人体动作识别方法.pdf

176201753（4）ComputerEngineeringandApplications计算机工程与应用强鲁棒性的可穿戴传感器的人体动作识别方法刘锦怡1张乐12胡海波12朱贺3LIUJinyi1ZHANGLe12HUHaibo12ZUHe31.重庆大学软件学院重庆4013312.信息物理社会-可信服务计算教育部重点实验室重庆4000443.重庆大学通信工程学院重庆4000441.SchoolofSoftwareEngineeringChongqingUniversityChongqing401331China2.KeyLaboratoryofDependableServiceComputinginCyberPhysicalSocietyMinistryofEducationChongqing400044China3.SchoolofTelecommunicationEngineeringChongqingUniversityChongqing400044ChinaLIUJinyiZHANGLeHUHaiboetal.Strongrobustnesshumanactivityrecognitionbasedonwearablesensors.ComputerEngineeringandApplications201753（4）：176-183.Abstract：Humandailyactivityrecognitionusingmobilepersonalsensingtechnologyplaysacentralroleinthefieldofpervasivehealthcare.Inthispaperanovelhumanactivityrecognitionframeworkispresentedtoreducetheimpactofsen-sorsdisplacement.Itutilizeshigh-precisionsensortocapturesignal.AccordingtoperiodicfeaturesofhumanmovementthecorrespondingfrequencydomainisgotbyFastFourierTransform.Theprincipalcomponentsanalysisisusedtoextractcompositeindicator.Afterextendprocesstheinputdataaself-organizingneuralnetworkmodelsisbuiltforgesturerecog-nition.Experimentalresultsdemonstratetheeffectivenessoftheschemeandinidealconditionstheaccuracyofcertainrelationshipcanget97.5%.Keywords：fastFouriertransform;self-organizationneuralnetwork;principalcomponentanalysis;humanactivityrecognition摘要：为了降低可穿戴传感器在传感器移位时对动作识别率的影响对可穿戴传感器的动作识别进行了研究。采用高精度传感器采集不同部位的输出信号根据运动的周期特点对输出信号进行去噪和快速傅里叶变换将其转化为频域信号。再使用主成分分析法提取综合指标并对自组织神经网络进行训练实现动作识别。最差情况下识别准确率可达到92.0%较好情况下甚至可达到97.5%传感器移位情况下的识别率甚至更高。关键词：快速傅里叶变换；自组织神经网络；主成分分析；动作识别文献标志码：A中图分类号：TP391.4doi：10.3778/j.issn.1002-8331.1507-0126近年来随着电子通信技术的发展和传感器精度的率同时也不方便在户外以及监视死角的应用。微型提高可穿