20.500.12556/DKUM-57822 REALNOČASOVNA METODA SLEDENJA OBRAZA V DIGITALNEM ZAPOREDJU SLIK NA PODLAGI INFORMACIJ O BARVI KOŽE IN PRIMERJAVA Z METODO VIOLA-JONES REAL-TIME FACE TRACKING METHOD IN A SEQUENCE OF DIGITAL IMAGES ON THE BASIS OF SKIN COLOUR INFORMATION AND COMPARISON WITH VIOLA-JONES METHOD V diplomskem delu smo se ukvarjali z realnočasovno metodo sledenja obraza v digitalnem zaporedju slik na podlagi barve kože. Predstavili smo obstoječe metode za lokaliziranje obrazov. Izdelali in preizkusili smo metodo, ki v realnem času poišče obrazno regijo in ji sledi skozi zaporedje slik. V metodi najprej podvzorčimo slike, s čimer zmanjšamo časovno zahtevnost metode, sledi iskanje področja, kjer je prišlo do spremembe na slikah iz zaporedja. Zatem poiščemo potencialne obrazne regije, jih dodatno obdelamo, označimo in prikažemo. Delovanje metode smo preverili z 20 testnimi skupinami slik, ki smo jih zajeli med eksperimentom. Natančnost zaznavanja smo primerjali z metodo Viola-Jones. Eksperimentalni rezultati so pokazali, da je naš algoritem glede na metrike, ki smo jih uporabili, uspešen. Naša metoda je za obdelavo ene slike potrebovala v povprečju 37,72 ms, medtem ko je za obdelavo ene slike metoda Viola-Jones potrebovala 79,81 ms. Dobljeni rezultati so se pokazali kot zelo obetavni. Z izboljšavami bi lahko bila naša metoda uporabna tudi v realnih situacijah. In this thesis, we deal with real-time face tracking method in a digital image sequence based on the skin colour. We survey the existing methods for the localization of faces. We implemented and tested a method that searches and follows the facial region through a sequence of images in real time. The method sub-samples the images first, which reduces the method’s time complexity, and searches for the changes in image sequence afterwards. Found potential facial region are pre-processed, labelled, and displayed. The method’s efficiency was checked by 20 groups of test images captured in the experiment. Its accuracy was compared to the Viola-Jones method. Experimental results showed that our algorithm produces satisfactory detections according to the metrics used. The processing time of our method yields 37.72 ms for one image, on average, while the processing of one image by the Viola-Jones method requires 79.81 ms. The results obtained have proved very promising. With improvements, our method can also be used in real situations. digitalna obdelava slik zaporedje slik zaznava obrazov sledenje obrazu barva kože algoritem Viola-Jones digital image processing image sequence face detection face tracking skin colour algorithm Viola-Jones true true false [M. Frangež] Slovenski jezik Angleški jezik Diplomsko delo 2016-03-18 17:25:54 2016-04-07 12:50:11 2022-06-15 03:09:07 2019-07-11 10:19:28 2016 0 Maribor 0 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 004.932.021:004.031.43(043.2) 19488022 URN:SI:UM:DK:EBMPP0SW UN_Frangez_Mitja_2016.pdf UN_Frangez_Mitja_2016.pdf 1 9661DEED2730A90BA79A4F8A278FFE4A 6f8ebd7cf2f8d6daba827e94b63780b8977d9dad66f4b5ec81f22004dfbc2cae fc73a6e7-7c0b-11eb-bb7a-00155d0001ca https://dk.um.si/Dokument.php?lang=slv&id=87386 Fakulteta za elektrotehniko, računalništvo in informatiko 0 0 0