Trt lt, p gchtrng mch, nh bng theo yu cu k thut. Talking to reporters of Labour newspaper soon after receiving information on the L gi tr vt liu chnh, vt liu ph, cc cu kin hoc cc b phn ri l, vt liu lun chuyn khng k vt liu ph cn dng cho my mc, phng tin vn chuyn v nhng vt liu tnh trong chi ph chung cn cho vic thc hin v hon thnh khi lng cng tc xy lp.
- Published on
26-Oct-2015 - View
183 - Download
6
![Lun Cng Vn Tt Pdf Lun Cng Vn Tt Pdf](https://reader001.docslide.net/reader001/html5/20170912/563dba7f550346aa9aa62a68/bg2.png)
Embed Size (px)
DESCRIPTION
OpenCV
Transcript
![Lun Cng Vn Tt Pdf Lun Cng Vn Tt Pdf](https://img.dokumen.tips/img/1200x630/reader008/image/20181030/55cf9aa9550346d033a2cd65.png)
- I HC QUC GIA TP. H CH MINHTRNG I HC BCH KHOANG TH THU HOATHEO VT I TNG S DNG MIXTURE OF GAUSSIAN MODEL V PARTICLE FILTER(Object tracking based on Mixture of Gaussian Model andParticle Filter)CNG LUN VN THC STP. H CH MINH 2013
- I HC QUC GIA TP. H CH MINHTRNG I HC BCH KHOANG TH THU HOATHEO VT I TNG S DNG MIXTURE OF GAUSSIAN MODEL V PARTICLE FILTERCHUYN NGNH: KHOA HC MY TNHM S CHUYN NGNH: 60.48.01CNG LUN VN THC SHNG DN KHOA HCTS. NGUYN THANH BNHTP. H CH MINH - 2013
- cng lun vn thc s1MC LCM U................................................................................................................... 2ng lc nghin cu, cc thch thc .................................................................. 2B cc ca ti ................................................................................................ 3NI DUNG ............................................................................................................... 4CHNG 1- GII THIU...................................................................................... 41.1 Gii thiu ti.......................................................................................... 41.2 Ni dung ti........................................................................................... 5Pht biu bi ton ....................................................................................... 5Gii hn ti............................................................................................ 51.3 Mc tiu ti............................................................................................ 51.4 Phng php nghin cu ............................................................................ 6CHNG 2 CNG TRNH NGHIN CU LIN QUAN ................................ 72.1 Gii thiu v cc gii thut ......................................................................... 72.2 Cc cng trnh nghin cu lin quan........................................................... 9CHNG 3 - BI TON THEO VT I TNG V HNG TIP CN 113.1 Qu trnh pht hin v theo vt i tng ................................................. 113.1.1 Pht hin i tng chuyn ng (Moving object detection) ............ 123.1.2 M hnh i tng (Object Modeling).............................................. 133.2 Gii thut xut ..................................................................................... 173.2.1 Object Extraction from background.................................................. 173.2.2 Object Tracking................................................................................ 19CHNG 4- KT QU D KIN T C................................................. 234.1 Kt qu d kin ........................................................................................ 234.2 Phng php nh gi kt qu .................................................................. 234.3 Nhng ng gp ca nghin cu .............................................................. 23D KIN K HOCH THC HIN................................................................... 24TI LIU THAM KHO
- cng lun vn thc s2M U ng lc nghin cu, cc thch thcTheo vt i tng (Object Tracking) l bi ton thuc lnh vc th gic my tnh. Trong mi nm tr li y, cng vi tc pht trin ca khoa hc k thut, con ngi cng c nhu cu s dng cc h thng thng minh vi mc t ng ha ngy cng cao. Mt s ng dng ca lnh vc th gic my tnh bao gm hthng: kim sot quy trnh (trong lnh vc robot), iu hng (trong giao thng v robot), pht hin s kin (an ninh v gim st), m hnh ha i tng (phn tch nh y khoa), gim st t ng (trong cc ng dng sn xut).Trong lnh vc an ninh-gim st (security and surveillance), th gic my tnh c ng dng rt nhiu. H thng gim st (Surveillance system) bao gm ba quy trnh : Xc nh i tng (Object extraction), theo vt i tng (Object tracking) v nhn dng hnh vi (Action recognition). T lu tr thng tin thu thp c vo c s d liu hoc pht hin bt thng a ra cnh bo kp thi.S quy trnh ca h thng gim stTheo vt i tng trong video c th nh ngha l bi ton xc nh v tr cai tng theo thi gian khi i tng chuyn ng. Ty vo tng ng dng c thm b theo vt i tng (Tracker) cung cp cc thng tin khc nhau v i tng nh hnh dng, din tch, ta trung tm, hng chuyn ng, t c tha ra d bo v v tr di chuyn tip theo ca i tng hoc nhn dng hnh vi a ra cnh bo cho nhng hnh ng bt thng.Thu thp hnh nhPht hin i tngTheo vt i tngNhn dng hnh viLu trthng tinCnh bo
- cng lun vn thc s3Bi ton theo vt i tng l bi ton phc tp v trong video quan st c thxut hin cc vn :- Nhiu do phn gii ca camera thp, do iu kin khch quan (thi tit, k thut ghi hnh, nh sng)- i tng c chuyn ng phc tp, tc nhanh.- i tng c kch thc thay i, b che khut bi i tng khc - S thay i ca chiu sng, gc chiu sng- i tng c mu sc ging vi cnh nn.- i tng di chuyn khi vng quan st v xut hin tr liNgoi ra, yu cu theo vt n i tng hoc a i tng, hnh nh thu thp tmt hoc nhiu camera, yu cu x l thi gian thc cng l nhng thch thc ln trong bi ton theo vt i tng.V vy, theo vt i tng l lnh vc vn c cc nh khoa hc quan tm nghin cu.B cc ca tiLun vn chia thnh 4 chng:- Chng 1: Gii thiu v ti v ni dung s nghin cu.- Chng 2: Tng quan cc gii thut c xut, cc cng trnh nghincu lin quan n ti.- Chng 3: Trnh by cc hng tip cn phn tch v gii quyt bi tontheo vt i tng. Cc gii thut xut cng s c trnh by trong chng ny.- Chng 4: S d kin kt qu t c, nh gi kt qu v qua nu ln nhng ng gp ca ti nghin cu.
- cng lun vn thc s4NI DUNG CHNG 1- GII THIUChng mt s gii thiu v vn , mc tiu v ni dung nghin cu ca ti, gii hn ca ti v phng php nghin cu.1.1 Gii thiu tiHiu mt cch n gin, theo vt i tng l bi ton xc nh ta ca i tng ti mi khung hnh (frame) trong on video quan st khi i tng chuyn ng.Mt vi ng dng quan trng ca bi ton theo vt i tng nh:- Gim st t ng (Automated video surveillance): trong nhng ng dng ny h thng th gic my tnh c thit k kim sot (monitor) nhng chuyn ng trong mt vng (area), xc nh i tng chuyn ng v cnh bo khi thy bt k tnh hung kh nghi no. i hi h thng phi mnh phn bit c cc thc th t nhin v con ngi.- Robot vision: vi robot t ng, h thng iu hng (navigation) cn phi nhn bit c chng ngi vt (obstacle) trn ng i. V nu l nhng i tng di chuyn, robot cn kch hot h thng theo vt thi gian thc trnh va chm.- iu phi giao thng (traffic monitoring): Trn cc i l hoc cc trc ng chnh, giao thng c gim st lin tc qua camera. Bt k phng tin no vi phm lut giao thng hoc lin quan n nhng hnh vi phm php khc u d dng c pht hin nu h thng gim st c tch hp tnh nng theo vt i tng.- Animation: gii thut theo vt c th s dng m rng k thut lm phim hot hnh- Ngoi ra cn nhng ng dng trong motion-based recognition, videoindexing, human-computer interactionKhi xem xt bi ton theo vt i tng cn quan tm n cch biu din i tng (object representation), la chn c trng ph hp (feature selecton), m hnh ha i tng v chuyn ng ca i tng da trn cc c trng. C nhiu phng
- cng lun vn thc sphp c xut gii quyt bi ton theo vt i tng. Ty vo mi trng quan st, ng cnh, mc tiu quan st m la chn cc gii thut khc nhau.1.2 Ni dung tiVn t ra l lm sao t mt on video quan st, ta xc nh c u l i tng ang chuyn ng, theo di s di chuyn ca i tng v xy dng quo chuyn ng ca i tng.Pht biu bi tonCho trc tp d liu l on video cha i tng cn theo vtD liu u vo (input): on video cha i tng ang chuyn ng.D liu u ra (output): s qu o chuyn ng ca i tnginput outputGii hn tiNh phn tch trong phn m u, c nhiu thch thc trong bi ton theovt i tng khin cho bi ton tr nn rt phc tp. V vy, mi gii thut xut u km theo nhng gi thit quy nh nhng iu kin rng buc nht nh. Trong nghin cu ny lun vn ch xc nh i tng l con ngi, d liu t mt camera, v quan st c thc hin trong iu kin nh sng tt.1.3 Mc tiu tiMc tiu nghin cu l tm hiu cc kin thc c lin quan n h thng gim st, tm hiu v cc gii thut theo vt i tng, xy dng c mt gii thut hiu qu. C th, pht hin c i tng chuyn ng, phn tch i tng khi cnh nn v i tng khc, xc nh ta ca i tng trong mi khung hnh, lin kt cc ta c c qu o chuyn ng ca i tng.Gii thut theo vt i tng
- cng lun vn thc s61.4 Phng php nghin cuLun vn s i t vic tham kho cc cng trnh nghin cu trc y linquan n bi ton theo vt i tng- Xem xt cc gii thut tc gi s dng- Phn tch cc gii thut theo tng giai on- Tng hp v phn loi thut ton da trn cch la chn c trng v biudin i tng- nh gi u im ca tng thut ton cng nh nhng hn ch cn tn tiT la chn thut ton hiu qu nht ti mi giai on, kt hp cc thut ton xy dng nn mt gii thut gii quyt bi ton theo vt i tng trong nhng iu kin rng buc nu trn.Hin thc gii thut bng cng c Matlab. So snh kt qu t c vi kt qu ca cc cng trnh nghin cu trc nh gi mc hiu qu ca gii thut.Kt lun chng 1:Chng 1 nu ln cc ng dng ca h thng theo vt i tng, trnh by vni dung nghin cu, mc tiu v phng php nghin cu.
- cng lun vn thc s7CHNG 2 CNG TRNH NGHIN CU LIN QUANChng hai s tng hp mt s phng php nghin cu v trnh by mt vinghin cu lin quan n ti2.1 Gii thiu v cc gii thutBi ton theo vt i tng t ra nhiu vn cn xem xt khi tm kim gii thut. Nh mc tiu l con ngi hay phng tin? Theo vt n i tng hay a i tng? Mi trng trong nh hay ngoi tri? ng dng vi mc ch pht hin hnh vi bt thng hay ng dng theo vt trong cnh quay thi u trong th thao?V ng dng rng ri ca bi ton m c rt nhiu nh nghin cu xut v pht trin cc gii thut khc nhau. [1] phn chia cc k thut theo vt i tng thnh 4 dng:- Theo vt da trn vng i tng (Tracking based on a moving object region)Gii thut ny ch yu da vo thuc tnh ca blob nh kch thc, mu sc, hnh dng, vn tc (velocity), trng tm (centroid). u im ca gii thut l thi gian tnh ton nhanh v hiu qu vi s lng i tng t. Hn ch ca gii thut l khng hiu qu khi i tng b che khut bi i tng khc trong trng hp nhiu i tng.- Theo vt da trn ng nt ni bt ca i tng (Tracking based on an active contour of a moving object)Contour ca i tng c biu din bi mt snake. Gii thut ch yu da trn boundary ca i tng. u im l c hiu qu trong trng hp theo vt ngi i b (pedestrian) bng cch lc chn ng nt ca u; c th ci thin thi gian tnh ton. Hn ch l khng gii quyt c bi ton i tng b che khut mt phn (partial occlusion) v nu i tng b che khut hoc hai i tng chng lp ln nhau mt phn trong qu trnh khi to (tc l nhng frame u tin) th sgy ra li.- Theo vt da trn m hnh ha i tng (Tracking based on moving object model)M hnh ca i tng thng c quy v m hnh hnh hc ca i tng trong khng gian 3D v gii thut s nh ngha tham s xc nh i tng. Gii thut ny gii quyt c bi ton che khut mt phn nhng li nh hng n thi gian
- cng lun vn thc s8x l. u im ca gii thut l c chnh xc cao khi s lng i tng khng nhiu.- Theo vt da trn xc nh c trng ca i tng(Tracking based on selected features of moving object)La chn nhng c trng tiu biu ca i tng v xem xt cc c trng qua cc frame lin tip xc nh i tng di chuyn v theo vt. Khi i tng bche khut, mt hoc hai c trng khng th s dng, vn c th da vo mt trong nhng c trng cn li. Tuy nhin, li ny sinh bi ton gom cm c trng (feature clustering), lm sao xc nh c nhng c trng no l thuc cng mt i tng trong sut qu trnh theo vt (trng hp theo vt nhiu i tng). Trong [2] theo vt i tng c phn loi thnh ba phng php:- Theo vt da trn im (Point tracking)i tng c biu din bng tp cc im v cc im ny c lin kt da trn cc rng buc v chuyn ng, v tr ca i tng. Hn ch ca phng php l cn c mt c ch bn ngoi pht hin i tng trong mi frame.Gii thut tiu biu l Kalman Filter, Particle Filter, Multi Hypothesis Tracking [3]- Theo vt da trn nhn (Kernel tracking)M hnh ca i tng c th c biu din di dng mu (template), hoc m hnh mt (density based model) v d nh histogram. Theo vt c thc hin bng cch tnh ton chuyn ng ca i tng qua cc frame lin tip.Gii thut tiu biu l Mean-shift, Simple Template Matching, Support Vector Machine (SVM) [3]- Theo vt da trn hnh chiu (Silhouette tracking)Sau khi c lng vng i tng (Object region) trong mi frame, i tng c theo vt bng cch s dng thng tin m ha trong vng i tng. Cc thng tin ny c th di hnh thc l m hnh v hnh dng hoc mt ca i tng. Khi c m hnh i tng, theo vt c thc hin bng phng php so khp hnh dng (shape matching) hoc m rng ng vin (contour evolutions)Tiu biu l Contour Tracking, Shape Matching. [3]
- cng lun vn thc s9Hnh 2.1 Cc gii thut theo vt i tng [2,3]2.2 Cc cng trnh nghin cu lin quan Object Classification and Tracking in Video Surveillance [1]Qi Zang and Reinhard KletteH thng theo vt i tng c xy dng cho ng dng trong gim st giao thng (traffic surveillance)giai on u, s dng gii thut tr nn phn tch i tng, Mi im nh nn (background pixel) s c m hnh ha bng phn phi mixture of Gaussian. Giai on hai, gn nhn cho tng vng i tng (object region) v xc nh cc c trng: bouding rectangle (hnh ch nht nh nht cha i tng), color (khng gian mu RGB), center (trng tm ca hnh ch nht), velocity (spixel di chuyn/giy theo c 2 hng dc ngang). S dng SUSAN (b pht hin gc) xc nh gc ca phng tin trong mi bounding box. S dng phng php lai (hybrid method) kt hp Kalman Filter vi k thut so khp (matching) theo vt i tng.u im ca gii thut l gim c thi gian tnh ton khi s dng b pht hin gc trong vng bounding rectangle. V s dng t s cao/rng trong thng tin gc phn lp i tng l ngi i b hay phng tin, nhng ch c hiu qunu cc vng i tng l tch bit.Object TrackingPointTrackingKernelTrackingSilhouetteTrackingKalman FilterParticle FilterMulti Hypothesis TrackingMean-shiftSimple Template MatchingSupport Vector MachineContour TrackingShape Matching
- cng lun vn thc s10Adaptive meanshift for automated multi object tracking [4]C. Beyan A. Temizela ra b theo vt a i tng hon ton t ng da trn gii thut mean-shift. S dng Gaussian loi nhiu, bng v rt trch foreground. ng thi Gaussian xc nh bouding box, dng nh mt mt n nhn (kernel mask) gim vng tm kim v d bo v tr mi ca i tng.u im l pht hin c khi i tng vo hoc ra khi vng quan st. Cp nht b theo vt vi thng tin foreground ci tin mean-shift, lm cho gii thut c hiu qu c trong trng hp i tng thay i v hnh dng, kch thc. Tuy nhin, ch p dng vi trng hp camera tnh (static camera) Object tracking in an outdoor environment using fusion of features andcamera [5]Quming Zhou, J.K. AggarwalBi bo a ra mt h thng theo vt v phn lp i tng chuyn ng sdng mt hoc nhiu camera trong mi trng ngoi tri (outdoor). Kt hp cc c trng nh v tr, hnh dng, mu sc tng hiu qu theo vt i tng. Kt hp thng tin t cc camera c c qu o chuyn ng ca i tng. ng thi, gii quyt bi ton che khut bng cch s dng b lc Kalman m rng (extended Kalman Filter-EKF). Gii thut cng phn lp i tng thnh ba nhm: mt ngi (single person), nhm ngi (people group) v phng tin (vehicle). Tuy nhin EKF khng thnh cng nu i tng b che khut c 2 camera.Kt lun chng 2:Chng 2 tng hp cc phng php theo vt i tng theo mt s nghin cu trc y, nu nhng c im cng nh u, nhc im ca cc phng php ; trnh by tng quan v mt s nghin cu lin quan n theo vt i tng.
- cng lun vn thc s11inputoutputTrajectoryCHNG 3 - BI TON THEO VT I TNG V HNG TIP CNChng ba trnh by quy trnh tng bc theo vt i tng, mt s gii thut thng c p dng. Cui cng, nu m t c th v gii thut xut. 3.1 Qu trnh pht hin v theo vt i tngT d liu u vo l on video, qu trnh theo vt i tng bao gm cc bc:- Tch frame: Tch on video thnh cc frame nh.- Tr nn: X l cc frame xc nh cnh nn (background) v i tng. - Tin x l: Kh bng, nhiu v phn tch i tng khi cnh nn.- Pht hin i tng: Nhn dng i tng chuyn ng, biu din i tngbng cc c trng.- Theo vt i tng: Xc dnh v tr ca i tng ti tng frame.Hnh 3.1 S qu trnh theo vt i tngvideoTch Frame (Image Frame)Tr nn (Background Subtraction)Pht...