Realistic 3D Virtual Fitting Room
Kuupäev
2018-04-24
Autorid
Ajakirja pealkiri
Ajakirja ISSN
Köite pealkiri
Kirjastaja
Abstrakt
Virtuaalne riiete proovimine on üks põhilistest teenustest, mille pakkumine võib suurendada rõivapoodide edukust, sest tänu sellele lahendusele väheneb füüsilise töö vajadus proovimise faasis ning riiete proovimine muutub kasutaja jaoks mugavamaks. Samas pole enamikel varem välja pakutud masinnägemise ja graafika meetoditel õnnestunud inimkeha realistlik modelleerimine, eriti terve keha 3D modelleerimine, mis vajab suurt kogust andmeid ja palju arvutuslikku ressurssi. Varasemad katsed on ebaõnnestunud põhiliselt seetõttu, et ei ole suudetud korralikult arvesse võtta samaaegseid muutusi keha pinnal. Lisaks pole varasemad meetodid enamasti suutnud kujutiste liikumisi realistlikult reaalajas visualiseerida. Käesolev projekt kavatseb kõrvaldada eelmainitud puudused nii, et rahuldada virtuaalse proovikabiini vajadusi. Välja pakutud meetod seisneb nii kasutaja keha kui ka riiete skaneerimises, analüüsimises, modelleerimises, mõõtmete arvutamises, orientiiride paigutamises, mannekeenidelt võetud 3D visuaalsete andmete segmenteerimises ning riiete mudeli paigutamises ja visualiseerimises kasutaja kehal. Selle projekti käigus koguti visuaalseid andmeid kasutades 3D laserskannerit ja Kinecti optilist kaamerat ning koostati nendest andmebaas. Neid andmeid kasutati välja töötatud algoritmide testimiseks, mis peamiselt tegelevad riiete realistliku visuaalse kujutamisega inimkehal ja suuruse pakkumise süsteemi täiendamisega virtuaalse proovikabiini kontekstis.
Virtual fitting constitutes a fundamental element of the developments expected to rise the commercial prosperity of online garment retailers to a new level, as it is expected to reduce the load of the manual labor and physical efforts required. Nevertheless, most of the previously proposed computer vision and graphics methods have failed to accurately and realistically model the human body, especially, when it comes to the 3D modeling of the whole human body. The failure is largely related to the huge data and calculations required, which in reality is caused mainly by inability to properly account for the simultaneous variations in the body surface. In addition, most of the foregoing techniques cannot render realistic movement representations in real-time. This project intends to overcome the aforementioned shortcomings so as to satisfy the requirements of a virtual fitting room. The proposed methodology consists in scanning and performing some specific analyses of both the user's body and the prospective garment to be virtually fitted, modeling, extracting measurements and assigning reference points on them, and segmenting the 3D visual data imported from the mannequins. Finally, superimposing, adopting and depicting the resulting garment model on the user's body. The project is intended to gather sufficient amounts of visual data using a 3D laser scanner and the Kinect optical camera, to manage it in form of a usable database, in order to experimentally implement the algorithms devised. The latter will provide a realistic visual representation of the garment on the body, and enhance the size-advisor system in the context of the virtual fitting room under study.
Virtual fitting constitutes a fundamental element of the developments expected to rise the commercial prosperity of online garment retailers to a new level, as it is expected to reduce the load of the manual labor and physical efforts required. Nevertheless, most of the previously proposed computer vision and graphics methods have failed to accurately and realistically model the human body, especially, when it comes to the 3D modeling of the whole human body. The failure is largely related to the huge data and calculations required, which in reality is caused mainly by inability to properly account for the simultaneous variations in the body surface. In addition, most of the foregoing techniques cannot render realistic movement representations in real-time. This project intends to overcome the aforementioned shortcomings so as to satisfy the requirements of a virtual fitting room. The proposed methodology consists in scanning and performing some specific analyses of both the user's body and the prospective garment to be virtually fitted, modeling, extracting measurements and assigning reference points on them, and segmenting the 3D visual data imported from the mannequins. Finally, superimposing, adopting and depicting the resulting garment model on the user's body. The project is intended to gather sufficient amounts of visual data using a 3D laser scanner and the Kinect optical camera, to manage it in form of a usable database, in order to experimentally implement the algorithms devised. The latter will provide a realistic visual representation of the garment on the body, and enhance the size-advisor system in the context of the virtual fitting room under study.
Kirjeldus
Väitekirja elektrooniline versioon ei sisalda publikatsioone
Märksõnad
human, body measures, three-dimensional graphics, 3D graphics, computer modelling, visualization, algorithms