[1]郭迎春,郑婧然,于 洋.基于TBGC的航拍视频车辆检测算法 [J].河北工业大学学报,2019,48(04):8-18.[doi:10.14081/j.cnki.hgdxb.2019.04.002]
 GUO Yingchun,ZHENG Jingran,YU Yang.Vehicle detection algorithm based on TBGC in aerial video [J].Journal of Hebei University of Technology,2019,48(04):8-18.[doi:10.14081/j.cnki.hgdxb.2019.04.002]
点击复制

基于TBGC的航拍视频车辆检测算法
()
分享到:

《河北工业大学学报》[ISSN:1007-2373/CN:13-1208/T]

卷:
48
期数:
2019年04期
页码:
8-18
栏目:
计算机应用
出版日期:
2019-08-25

文章信息/Info

Title:
Vehicle detection algorithm based on TBGC in aerial video
文章编号:
1007-2373(2019)04-0008-11
作者:
河北工业大学 人工智能与数据科学学院
Author(s):
GUO Yingchun ZHENG Jingran YU Yang
School of Artificial Intelligence, Hebei University of Technology
Keywords:
aerialvideovehicledetectionsupportvectormachineTBGCfeatures
分类号:
TP391.41
DOI:
10.14081/j.cnki.hgdxb.2019.04.002
文献标志码:
A
摘要:
针对移动航拍视频中车辆检测准确度低的问题,提出一种基于三邻域点二值梯度轮廓(Three-neigh? bor-pointBinaryGradientContour,TBGC)特征的航拍车辆检测算法。对相邻帧图像进行SURF(speeded-upro? bustfeatures)特征点提取匹配,利用角度判别剔除错误匹配点完成图像配准,采用帧间差分获得运动目标 的候选区域。由于传统二值梯度轮廓(BinaryGradientContours,BGC)特征忽略中心像素特性,提出基于3×3 邻域相邻像素点量化操作的TBGC特征。提取候选区域的TBGC特征,并利用支持向量机(SupportVector Machine,SVM)完成最终的航拍视频车辆检测。实验中利用提出的TBGC特征在8个数据集上分别与BGC1、 LBP、HOG特征进行对比实验,实验结果表明TBGC算法的检测率明显优于传统经典算法,平均检测率为 93.09%,并且具有较好的鲁棒性。
Abstract:
Toimprovethevehicledetectionprecisioninaerialvideo,avehicledetectionmethodforaerialvideoispro? posedbasedonthree-neighbor-pointbinarygradientcontour(TBGC)features.First,theextractionofspeed-uprobust features(SURF)wasextractedandmatchedforadjacentframeimages,anderror-matchingpointswereeliminatedbyan? glediscriminationtorealizeimageregistration.Then,thecandidateregionsofthemovingtargetwereobtainedaccording totheframedifference.Sincethetraditionalbinarygradientcontour(BGC)featuresignorethepixelcharacteristicsatthe center,theTBGCfeaturesbasedonthecomparisonofadjacentpixelpointsina3×3neighborhoodwereproposed.In thisway,theTBGCfeaturesincandidateregionswereextracted,andsupportvectormachine(SVM)wasusedtocom? pletethevehicledetectioninaerialvideoatlast.Intheexperiment,theproposedTBGCfeatureswerecomparedwithfea? turesofBGC1,LBP,andHOGoneightdatasetsrespectively.ExperimentalresultsshowthatthedetectionrateofTBGC algorithmisobviouslybetterthanthetraditionalclassicalalgorithms,withanaveragedetectionrateof93.09%andstron? gerrobustness.

参考文献/References:

[1] KUMARR.Aerialvideosurvellianceandexploitation[M]//Video-BasedSurveillanceSystems.Springer,Boston,MA,2002:29-38. [2] 刘亚伟,李小民.无人机航拍视频中目标检测和跟踪方法综述[J].飞航导弹,2016(9): 53-56,70. [3] XUYZ,YUGZ,WANGYP,etal.Ahybridvehicledetectionmethodbasedonviola-jonesandHOG+SVMfromUAVimages[J].Sensors,2016, 16(8):1325. [4] MADHOGARIAS,BAGGENSTOSSP,SCHIKORAM,etal.CardetectionbyfusionofHOGandcausalMRF[J].IEEETransactionsonAerospace andElectronicSystems,2015,51(1):575-590. [5] CHENZY,WANGC,LUOH,etal.Vehicledetectioninhigh-resolutionaerialimagesbasedonfastsparserepresentationclassificationandmulti? orderfeature[J].IEEETransactionsonIntelligentTransportationSystems,2016, 17(8):2296-2309. [6] LIUK,SKIBBEH,SCHMIDTT,etal.Rotation-invariantHOGdescriptorsusingfourieranalysisinpolarandsphericalcoordinates[J].International JournalofComputerVision,2014,106(3):342-364. [7] MORANDUZZOT,MELGANIF.DetectingcarsinUAVimageswithacatalog-basedapproach[J].IEEETransactionsonGeoscienceandRemote Sensing,2014,52(10):6356-6367. [8] 毛征,刘松松,张辉, 等.不同光照和姿态下的航拍车辆检测方法[J].北京工业大学学报,2016, 42(7):982-988. [9] MORANDUZZOT,MELGANIF.ASIFT-SVMmethodfordetectingcarsinUAVimages[C]//2012IEEEIternationalGeoscienceandRemoteSens? ingSymposium, 22-27July2012,Munich,Germany,2012:6868-6871[10] MORANDUZZOT,MELGANIF.Automaticcarcountingmethodforunmannedaerialvehicleimages[J].IEEETransactionsonGeoscienceandRe? moteSensing,2014, 52(3):1635-1647. [11] POOSTCHIM,PALANIAPPANK,SEETHARAMANG.Spatialpyramidcontext-awaremovingvehicledetectionandtrackinginurbanaerialimag? ery[C]//201714thIEEEInternationalConferenceonAdvancedVideoandSignalBasedSurveillance(AVSS),29Aug.-1Sept.2017,Lecce,Italy, 2017: 1-6. [12] XUYZ,YUGZ,WUXK,etal.Anenhancedviola-jonesvehicledetectionmethodfromunmannedaerialvehiclesimagery[J].IEEETransactions onIntelligentTransportationSystems,2017,18(7):1845-1856. [13] MALAGIVP,BABUDRR.Rotation-invariantfastfeaturebasedimageregistrationformotioncompensationinaerialimagesequences[C]//Pro? ceedingsofInternationalConferenceonCognitionandRecognition.Springer,Singapore,2018:211-221. [14] WANGGL,WANGXC,FANB,etal.Featureextractionbyrotation-invariantmatrixrepresentationforobjectdetectioninaerialimage[J].IEEE GeoscienceandRemoteSensingLetters,2017,14(6):851-855. [15] LINYD,HEHJ,YINZK,etal.Rotation-invariantobjectdetectioninremotesensingimagesbasedonradial-gradientangle[J].IEEEGeoscience andRemoteSensingLetters,2015, 12(4):746-750. [16] WANGGL,FANB,ZHOUZL,etal.Ordinalpyramidcodingforrotationinvariantfeatureextraction[J].Neurocomputing,2017,242:150-160. [17] AMMOURN,ALHICHRIH,BAZIY,etal.DeeplearningapproachforcardetectioninUAVimagery[J].RemoteSensing,2017, 9 (4):312. [18] CHENXY,XIANGSM,LIUCL,etal.Vehicledetectioninsatelliteimagesbyhybriddeepconvolutionalneuralnetworks[J].IEEEGeoscienceand RemoteSensingLetters,2014,11(10):1797-1801. [19] DENGZP,SUNH,ZHOUSL,etal.Towardfastandaccuratevehicledetectioninaerialimagesusingcoupledregion-basedconvolutionalneural networks[J].IEEEJournalofSelectedTopicsinAppliedEarthObservationsandRemoteSensing,2017, 10(8):3652-3664. [20] FERNÁNDEZA,ÁLVAREZMX,BianconiF.Imageclassificationwithbinarygradientcontours[J].OpticsandLasersinEngineering,2011,49(9/ 10):1177-1184. [21] BAYH,ESSATUYTELAARST,etalSpeeded-uprobustfeatures(SURF)[J].ComputerVisionandImageUnderstanding,2008,110(3):346359. [22] CHUMO,MATASJ,KITTLERJ.LocallyoptimizedRANSAC[C]//JointPatternRecognitionSymposium.Springer,Berlin,Heidelberg,2003:236243.

备注/Memo

备注/Memo:
收稿日期:2018-04-23 基金项目:天津市科技计划项目(15ZCZDNC00130);河北省自然科学基金(F2015202239);天津市科技计划项目(17ZLZDZF00040) 作者简介:郭迎春(1970—),女,副教授。通信作者:于洋(1981—),男,讲师,yuyang@scse.hebut.edu.cn。
更新日期/Last Update: 2019-10-01