20.500.12556/DKUM-80232 Primerjava postopkov za prepoznavanje oseb v nekontroliranem okolju diplomsko delo Comparison of procedures for person recognition in an uncontrolled environment V diplomskem delu smo se ukvarjali z analizo metod za prepoznavanje oseb v nekontroliranem okolju. Cilj naše raziskave je bil preštudirati in primerjati različne metode, nato pa na osnovi postavljenih kriterijev izbrati najprimernejšo za različne vrste aplikacij. Najprej smo opravili pregled področja ter identificirali obstoječe odprtokodne metode. Zatem smo preučili metrike za merjenje zmogljivosti postopkov za prepoznavo oseb v nekontroliranem okolju. Po opravljenem študiju smo izbrali tri metode za testiranje, in sicer metodo Ageitgey/face_recognition, SphereFace ter OpenBR. Za namen testiranja smo preučili in namestili javni podatkovni zbirki LFW in UTKFace, ki vsebujeta osebe v nekontroliranem okolju. V eksperimentalnem delu smo testirali posamezno metodo s pomočjo eksperimentalnih podatkovnih zbirk. Testiranje smo izvajali na prenosnem računalniškem sistemu, pri čemer smo beležili število pozitivnih in negativnih detekcij. Z dobljenimi rezultati smo nato izračunali šest uveljavljenih metrik na področju prepoznavanja oseb v nekontroliranem okolju, opazovali pa smo tudi procesorsko in prostorsko zahtevnost metod. Najslabše se je odrezala metoda Ageitgey/face_recognition, najbolje pa metoda SphereFace. Metoda OpenBR je bila na zbirki LFW celo enakovredna metodi SphereFace. Na osnovi eksperimentiranj zaključujemo, da je za uporabo v domačem okolju najbolje uporabiti metodo Ageitgey/face_recognition. Za uporabo v profesionalnem okolju pa se odločimo med metodama SphereFace in OpenBR, in sicer glede na sistemske vire, ki jih imamo na razpolago. In the bachelor's thesis we dealt with the analysis of methods for recognizing persons in an uncontrolled environment. The aim of our research was to study and compare different methods, and then, based on the set criteria, select the most suitable for different types of applications. We first reviewed the field and identified existing open source methods. We then examined metrics to measure performance of procedures for identifying persons in an uncontrolled environment. After the study, we selected three procedures for testing, namely the Ageitgey/face_recognition, SphereFace and OpenBR procedures. For testing purposes, we examined and installed public databases LFW and UTKFace that contain individuals in an uncontrolled environment. In the experimental part, we tested individual procedures with the help of experimental databases. Testing was performed on a portable computer system, recording the number of positive and negative results. With the obtained results, we then calculated six established metrics in the field of recognizing persons in an uncontrolled environment. Processor and spatial complexity of procedures were also observed. Ageitgey/face_recognition performed the worst, and SphereFace performed the best. OpenBR on the LFW database was even equivalent to SphereFace. Based on the experiments, we conclude that it is best to use Ageitgey/face_recognition for use in the home environment. For use in a professional environment, we choose between OpenBR and SphereFace, depending on the system resources we have at our disposal. prepoznava oseb nekontrolirano okolje primerjava metod računalniški vid person recognition uncontrolled environment method comparison computer vision. true true false [Ž. Oblak] Slovenski jezik Angleški jezik Diplomsko delo/naloga 2021-09-02 14:23:57 2021-10-18 12:37:46 2022-08-09 22:39:40 0000-00-00 00:00:00 2021 0 Maribor 0 Maribor X, 46 f. 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 004.932.72(043.2) 90520579 VS_Oblak_Zan_2021.pdf VS_Oblak_Zan_2021.pdf 1 A60ECD9D4ED8FB6DC49355AD9CFFF43C e2c8532f1fb6b40cc6015fa3c17b004bae9dfa8bed8ad84c61db7574a0c31c45 ba08a665-0be9-11ec-a776-00155d581298 20.500.12556/dkum/855efb82-2d64-4419-bef0-0bffbefd1a34 https://dk.um.si/Dokument.php?lang=slv&id=152953 Fakulteta za elektrotehniko, računalništvo in informatiko 0 0 0