20.500.12556/DKUM-39873 ZAKRIVANJE GIBAJOČIH SE OBJEKTOV V VIDEO POSNETKIH IN PRIMERJAVA Z NJIHOVIMI MODELI 4D MASKING THE MOVING OBJECTS IN VIDEOS AND COMPARISON WITH THEIR 4D MODELS V tem magistrskem delu se ukvarjamo z dvema povezanima problemoma, za katera iščemo rešitve s pomočjo obdelave slik in računalniškega vida. V prvem delu se ukvarjamo z zakrivanjem gibajočih se objektov na video posnetkih, v drugem delu pa razvijemo postopek, s katerim primerjamo računalniške modele objektov in fotografije njihovih realizacij. Vhodne podatke zajemamo kot videe ali fotografije z digitalno kamero. Na posneti sceni poiščemo s pomočjo računalniške analize na zaporedju slik objekte, ki se premikajo, in jih zakrijemo. Programske rešitve smo prototipno razvijali z orodjem MATALB in knjižnico OpenCV, naknadno pa smo jih prenesli še v programski jezik C++, tako da se lahko izvajajo v realnem času. Računalniške modele objektov smo privzeli po standardu IFC iz gradbene informatike, ki predstavljajo 3D model zgradbe. Če dodamo še terminski plan, pridemo do 4D modela. Ujemanje načrta z realno situacijo smo izvedli tako, da smo ravninske projekcije 3D modelov poravnali s fotografijami realizacije in iskali razlike. Pri preizkušanju smo zgradbe simulirali z lesenimi maketami. Pri programiranju prototipnih rešitev smo si pomagali z računalniško grafiko in standardom OpenGL ter knjižnico EmguCV rešitvami za računalniški vid. Uporabili smo okolje .NET in programski jezik C#. Realizirani postopek za zakrivanje gibajočih se objektov smo statistično ocenili s senzitivnostjo in preciznostjo, iskanje razhajanj med modelom in realno situacijo pa z maketami. In this master’s thesis, we are dealing with two related problems that can be solved by using image processing and computer vision. In the first part, we go for masking of moving object in videos. In the second part, we develop a procedure that compares computer models of object to the photos of real situations. We capture input data as a video or a photo by a digital camera. Recorded scenes were analyzed by computer algorithms to detect the moving objects in videos, which we mask afterwards. A software prototype was developed in MATLAB and the OpenCV library, which was, afterwards, ported to a real time version in the C++ programming language. We constructed computer models of object by following the example of the IFC standard that is known in the field of construction informatics. 4D models were created from a scheduled sequence of 3D models. Our goal was to detect differences between the scheduled model and the corresponding photo of real situation. We registered a planar projection of 3D models with the photos looked for the differences. For the verification purposes, we built wooden models and their 4D computer counterparts. A software prototype was developed based on the computer graphic principles and the OpenGL standard. The computer vision part was supported by the EmguCV library and developed in the .NET environment with the C# programming language. Final validation of derived approaches was performed statistically by computing sensitivity and precision for masking the moving objects in videos, and by simulating 4D-model comparisons with real situation on wooden models. računalniški vid poravnava slik morfološki operatorji 4D modeli zakrivanje objektov computer vision image registration morphological operators 4D models masking of objects true true false [U. Kolarič] Slovenski jezik Angleški jezik Magistrsko delo/naloga 2013-03-12 00:10:37 2013-03-22 11:38:10 2022-05-12 23:16:32 2021-04-22 03:32:46 2013 0 Maribor 0 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 004.925.2:621.397(043.3) 16823830 URN:SI:UM:DK:ES33UY2G MAG_Kolaric_Uros_2013.pdf MAG_Kolaric_Uros_2013.pdf 1 3E456C94CEF49C312C5FCD1807C97150 c6f532ce1cf72df9fb8414eeea38a612c586714473d3cb7a1b515e141343fbd8 3f3239aa-7c09-11eb-bb7a-00155d0001ca 20.500.12556/dkum/fd54230f-3c48-4c7b-b58f-c0283d9cab6b https://dk.um.si/Dokument.php?lang=slv&id=54408 Fakulteta za elektrotehniko, računalništvo in informatiko 0 0 0