基于视觉的选煤厂智能监控系统研究

《基于视觉的选煤厂智能监控系统研究》由会员上传分享，免费在线阅读，更多相关内容在教育资源-天天文库。

1、AbstractThecoalsecurityhasattractedmuchattentionrecently.Coaldressingisanessentialprocessincoalproduction,whereaccidentsoccurringfrequently.Thisthesisaddsvisualelementstotheexistingvideosurveillancesystemofcoalpreparationplant.Upgradeittoanintelligentmonitoringsystemwh

2、ichcancarryonhumanmovementdetection,trackindependently,andalarmwhensomethingabnormal,etc.Themiantaskofthisthesisincludes:a.Movingtargetdetection.Takeimagenoiseprocessingforvideoframe.It’sthroughthebackgroundsubtractionthatbasedonmixtureGaussianmodeltodetectmotiontarget

3、,thenaccomplishtargetsegmentationroughlyaccordingtotheproportionofhumanbody.b.Featureextractionofhumanimagebasedon6-LBPandHOG.Thisthesiselaboratesandmakesuseoftwofeatureextractionmethods,onesistheHOGfeatureswhichbasedontheedgegradientofappearanceandoutlineoflocaltarget

4、,andtheotheroneistheLBPfeaturewhichbasedonthecolorandtextureinformationoflocalarea.Anotherfeatureextractionmethod,e-LBPisproposedonthebasisofthetraditionalLBPfeatureextractionmethods,removepartoftheredundantinformationandretainmoreedgeprofileinformationofhumanbody.Fina

5、lly,usethePCAmethodtoreducethefeaturedimension.ExperimentsshowthatthefeatureextractionmethodbasedonHOGcombiningwith6-LBPfeaturehasgoodcharacterizationresultforthehumanbody.c.Thedesignofhumanclassifierviasparserepresentation.Consideringthe sparserepresentationtheoryhasd

6、iscriminationwhichcanchoosetheeffectivesubset andrejectinvalidsubset,thethesisusesparserepresentationmethodthatgetsthe sparsesolutionsvia-minimizationtodesignthehumanclassifier.Experiments showthatthehumanclassifiersbasedonsparserepresentationhasbettereffectthan suppor

7、tvectormachine(SVM)inlow-dimensionalfeature.Especiallywhenthe occlusionhappensitachievesasatisfiedeffect.d.CombinedCamShiftalgorithmwithKalmanpredictionformovinghuman tracking.CamShiftalgorithmisimprovedbasedontheMeanShiftalgorithm.Ithas theadvantagesofsimplecalculatio

8、nandstrongrobustness.ThroughtheKalmanfilter forecastthehumanmovementparameters,thisthesisovercomestracklossthatcaused