資源描述:
《基于均值偏移和粒子濾波的視頻目標(biāo)跟蹤算法研究》由會(huì)員上傳分享,免費(fèi)在線閱讀,更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫(kù)。
1、碩士論文基于均值偏移和粒子濾波的視頻目標(biāo)跟蹤算法研究摘要目標(biāo)跟蹤技術(shù)是視覺領(lǐng)域研究中的熱點(diǎn)問題,廣泛應(yīng)用于視頻監(jiān)控、生物醫(yī)學(xué)、導(dǎo)航制導(dǎo)等領(lǐng)域。而目標(biāo)運(yùn)動(dòng)過程中自身的形變、復(fù)雜背景的干擾以及各類噪聲、遮擋、光照等因素都會(huì)對(duì)目標(biāo)跟蹤產(chǎn)生影響。在復(fù)雜的現(xiàn)實(shí)環(huán)境下,有針對(duì)性地提高跟蹤算法的準(zhǔn)確性、魯棒性與實(shí)時(shí)性是目標(biāo)跟蹤研究的重點(diǎn)。本文就均值偏移算法和粒子濾波算法在目標(biāo)跟蹤中應(yīng)用,展開了如下研究:對(duì)于均值偏移算法,針對(duì)目標(biāo)尺度變化、遮擋及光線變化等情況提出了相應(yīng)的改進(jìn)方法。采取帶寬自適應(yīng)的改進(jìn)策略,有效地實(shí)現(xiàn)了對(duì)
2、尺度變化目標(biāo)的跟蹤;在LMS濾波和Kalman預(yù)測(cè)的基礎(chǔ)上,提出自適應(yīng)濾波MeanShift算法,增強(qiáng)了算法在目標(biāo)形變和遮擋情況下的魯棒性;引入基于HSV顏色模型的CamShifl算法,采用運(yùn)動(dòng)預(yù)測(cè)對(duì)搜索策略進(jìn)行改進(jìn),提高了算法應(yīng)對(duì)光線變化和目標(biāo)遮擋等情況的適應(yīng)性。針對(duì)粒子濾波算法,采用相關(guān)跟蹤和多特征融合的方法對(duì)其進(jìn)行優(yōu)化改進(jìn)。采用相關(guān)跟蹤策略,實(shí)現(xiàn)了基于粒子濾波的相關(guān)跟蹤方法,驗(yàn)證了粒子濾波算法的魯棒性;融合顏色特征和紋理特征建立觀測(cè)模型,同時(shí)在系統(tǒng)動(dòng)態(tài)模型中加入尺度變量,提出了一種多特征融合的粒子濾
3、波算法,提高了跟蹤算法的魯棒性。對(duì)均值偏移算法和粒子濾波算法進(jìn)行融合,在此基礎(chǔ)上對(duì)搜索策略進(jìn)行改進(jìn)。利用均值偏移算法的聚類作用,將粒子樣本收斂在更接近目標(biāo)的真實(shí)位置的區(qū)域,把均值偏移理論嵌入到粒子濾波算法中以實(shí)現(xiàn)對(duì)目標(biāo)的跟蹤;采用Bhattacharyya系數(shù)衡量目標(biāo)跟蹤狀況,結(jié)合搜索策略跟蹤目標(biāo),對(duì)均值偏移與粒子濾波融合算法進(jìn)行相應(yīng)的改進(jìn),改善了算法的跟蹤性能。關(guān)鍵詞:目標(biāo)跟蹤,均值偏移,粒子濾波器,卡爾曼濾波,最小均方濾波碩士論文AbstractTargettrackingisahotissueinv
4、isionfield,whichiswidelyusedinvideosurveillance,bio—medicine,inertialguidance,navigationsystemandSOon.W11ilethetargetismoving,thedisadvantagefactorshaveaseriousinfluenceonthetargettracking,suchasthedeformation,theinterferenceofthecomplexbackgroundandvario
5、ustypesofnoise,shelterandlight.Undertherealcomplexcondition,improvingtheaccuracy,robustnessandreal—timeperformanceofthetrackingalgorithmisimportantinthefieldoftargettracking.T11ispaperfocusesontheapplicationofMeanShiftalgorithmandParticleFilteralgorithmin
6、thetargettracking.111emainstudiesareshowedasfollows:Totheconditionofchangingtarget’Sscale,shelteredtargetandchanginglight,animprovedMeanShiftalgorithmisintroduced.Firstly,theadaptivebandwidthMeanShiftalgorithmisadoptedtotrackthetargeteffectivelywhosesizei
7、schanging.Secondly,anadaptivefilterMeanShiftalgorithmbasedonLMSfilterandKalmanpredictionalgorithmisputforward,whichenhancestherobustnessofthealgorithmundertheconditionofthedeformationtargetandtheshelteredtarget.Thirdly,CamShiftalgorithmbasedontheHSVcolorm
8、odelisintroduced.髓emotionpredictionisadoptedtoenhancetheadaptabilityofthetrackalgorithmundertheconditionofchanginglightandshelteredtarget.Correlativetrackingstrategyandmulti—featurefusionisadoptedtoimproveParticleFi