DEVELOPMENT OF COMPUTER VISION BASED APPLICATION FOR MOBILE PHONE IPHONE

Cilj diplomskega dela je razviti zabavno aplikacijo na osnovi metod računalniškega vida, ki se bo izvajala na mobilnem telefonu (tj. mobilnik) Apple iPhone. Zastavimo si temeljno vprašanje, ali lahko s pomočjo aplikacije, ki se izvaja na mobilniku, v skoraj realnem času razpoznamo dele človeškega telesa. Razvijemo aplikacijo iDelude, ki opravi obdelovanje po naslednjem postopku. S fotoaparatom mobilnika najprej posnamemo sliko osebe. Z metodami računalniškega vida in digitalne obdelave slik nato razpoznamo telo, ga obarvamo v prevladujočo kožno barvo osebe, intimne dele pa prekrijemo z nadomestnimi oblačili. Položaje telesa in telesnih struktur določimo na osnovi razpoznanega obraza. Obraz razpoznavamo s funkcijami iz odprto kodne knjižnice za računalniški vid, in sicer knjižnice OpenCV. Te funkcije v določenih delih ustrezno prilagodimo oz. včasih tudi nadgradimo. Velikost razpoznanega obraza služi za umestitev dodatnih oblačil na sliko, razpoznana prevladujoča barva obraza pa je barva, s katero nadomestimo obstoječa oblačila osebe. V diplomskem delu tudi opišemo vsa razvojna okolja, orodja in druga potrebna znanja, ki so nujna za razvoj mobilne aplikacije za mobilnik iPhone. Na osnovi rezultatov ugotovimo, da je implementacija takšne aplikacije možna, hkrati pa pokažemo, da so današnji mobilniki že dovolj zmogljivi, da lahko z njimi v doslednem času opravimo kombinacijo zahtevnejših operacij iz področja računalniškega vida.An aim of this diploma work is to develop an entertaining application based on computer vision methods which will run on mobile phone Apple iPhone. A fundamental question raised is if it’s possible to recognize human body parts in near real time with application running on the mobile phone. We develop the iDelude application that performs an image processing as follows. First, a picture of person is captured by using phone’s camera. Human body is then recognized by using computer vision and digital image processing methods. Afterwards, the recognized body is colored in person’s dominant skin color, while intimate body parts are covered by substitute clothing. A determination of body structure and postures is based on recognized face. The face is recognized by using functions from an open source library for computer vision, namely OpenCV library. We appropriately adjust and sometimes also upgrade some of these functions. The size of recognized face is used for overlaying substitute clothing on image. A recognized dominant face color is a color, which is used for replacing existing person’s clothing. In this diploma work, we are describing also development environments, tools and other necessary skills that are essential for the development of mobile application for the iPhone handset. Based upon results we conclude that the implementation of such application is possible and, at the same time, we prove that nowadays mobile phones are powerful enough to perform a combination of complex operations from the field of computer vision in nearly real time