20.500.12556/DKUM-45636 PREPOZNAVANJE ČUSTEV S POMOČJO SENZORJA KINECT, PROGRAMSKEGA ORODJA FACESHIFT IN ANALIZE GLAVNIH KOMPONENT EMOTION RECOGNITION WITH KINECT SENSOR, FACESHIFT SOFTWARE AND PRINCIPAL COMPONENT ANALYSIS V magistrski nalogi smo preučili učinkovitost računalniško-podprtega vrednotenja čustev pri ljudeh. V ta namen smo s senzorjem Kinect in programskim orodjem FaceShift posneli premike 121 obraznih točk med osnovnimi mimikami, kot so dvigovanje obrvi, zapiranje oči, vertikalno odpiranje ust, horizontalno odpiranje ust in napihovanje ličnic. Z analizo glavnih komponent smo preverili ponovljivost in komplementarnost posnetih obraznih mimik in za vsako obrazno mimiko izračunali projektor meritev v njihovo prvo glavno komponento. Sledila je klasifikacija obrazne mimike med izražanjem naslednjih čustev: nasmeh, začudenje, gnus in žalost. Meritve posameznega čustva smo najprej projicirali v prostor prvih glavnih komponent vseh petih obraznih mimik, nato pa smo omenjene projekcije klasificirali z metodo podpornih vektorjev. Rezultati potrjujejo zelo dobro ponovljivost in komplementarnost projektorjev zgoraj omenjenih obraznih mimik, ter tudi zelo dobro razpoznavo čustev. Dobro razpoznamo nasmeh, začudenje in gnus, slabše pa žalost. Slabše rezultate razpoznave žalosti pripisujemo slabi sledljivosti ustnim kotičkom v orodju FaceShift med izražanjem tega čustva. We studied the efficiency of computer-based recognition of human emotions from facial expressions. For this purpose, we recorded the movement of 121 facial markers with Kinect sensor and FaceShift software while expressing basic facial mimics like lifting the eyebrows, closing the eyes, opening the mouth vertically and horizontally and blowing cheeks up. For every facial mimic, we used principal component analysis to check the repeatability and the complementarity of the recorded measurements and calculated the projector of the measurements into their first principal component. These projectors were then used for dimensionality reduction in measurements of facial expressions of smile, astonishment, disgust and sadness. First, we projected the measurements of every emotion into the space of the first principal components of each facial mimic. Afterwards, we used support vector machine method to classify afore mentioned projections into different classes of emotions. The results confirm very good repeatability and complementarity of selected basic facial mimics and highly accurate emotion recognition. The smile, astonishment and disgust were recognized with accuracy of 100 %. The recognition of sadness was less accurate, mainly due to the problems with the tracking of the mouth corners in the FaceShift model. razpoznavanje čustev obrazne mimike analiza glavnih komponent sledenje obrazni mimiki orodje FaceShift emotion recognition facial mimics principal component analysis facial mimics tracking FaceShift software true true false [B. Brenčič] Slovenski jezik Angleški jezik Magistrsko delo/naloga 2014-08-31 02:27:27 2014-10-30 15:11:53 2022-05-16 03:05:53 2021-05-03 03:23:29 2014 0 Maribor 0 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 004.93'1:004.4'275(043.2) 18389782 URN:SI:UM:DK:8XAIWBUM MAG_Brencic_Bostjan_2014.pdf MAG_Brencic_Bostjan_2014.pdf 1 63878E1DF10E498D95280EFCFF22D7BB 358e772bf120746dae832aed1330ed562d4477cf48b52f016254881a13ef292c 20431ec9-7c0b-11eb-bb7a-00155d0001ca https://dk.um.si/Dokument.php?lang=slv&id=66205 Fakulteta za elektrotehniko, računalništvo in informatiko 0 0 0