| الوصف: |
V diplomskem delu smo se ukvarjali s študijem, preizkušanjem in implementiranjem osnovnih postopkov računalniške obdelave slik in računalniškega vida za razvoj avtomatiziranega snemanja rokometnih tekem, odigranih v dvorani. V okviru diplomskega dela smo razvili računalniški algoritem, ki omogoča povsem avtonomno snemanje dogodkov na rokometnem igrišču, ki jih je treba posneti s kamero. Z dvema na strop pritrjenima kamerama smo s pomočjo tehnik računalniške obdelave slik in vida izluščili igralce na igrišču. V nadaljevanju smo analizirali postavitev igralcev in ocenili, v kateri ciljni točki na igrišču se izvaja dogodek, ki ga je treba posneti. V eksperimentalnem delu diplomske naloge smo posneli video dejanske rokometne tekme, s pomočjo katere smo razvijali in testirali naš postopek. Testirali smo na petdesetih slikah, naključno izbranih z video posnetka, pri čemer smo bili pozorni, da je bila ciljna točka pravilno določena. Definirali smo območja, ki bi jih morala zajeti kamera za snemanje videa in jih primerjali z merjenim rezultatom. Rezultati so pokazali, da naša rešitev v 86 % pravilno zazna in posname pomemben dogodek rokometne tekme. Naš sistem je manj uspešen v izjemnih situacijah, kot so, na primer, protinapad ali met žoge na prazen gol. Na osnovi eksperimentalnih rezultatov trdimo, da naša rešitev dosega zadovoljivo uspešnost. In our diploma work we studied, tested and implemented the basic principles of digital image processing and computer vision to develop an automated recording of a handball match that is played in an arena. We developed an algorithm, that enables a fully automated recording of events on a handball field, that need to be recorded during the match. With two ceiling-based cameras and the help of digital image processing and computer vision techniques we extracted the players on the field. Next, we analyzed player positions and pinpointed the location and event that needed to be recorded. In the experimental part of our diploma work we recorded an actual handball match and used it to develop and test our process. We tested on 50 frames, that were randomly selected from the recording, if the target was pinpointed correctly. We manually defined the areas that the camera should cover and record, and compared them to the results of the automated recording. The results showed that our solution was correct in 86% of cases that needed to be detected and recorded in a handball match. Our system is less accurate in rarer situations such as counterattack and shooting on an empty goal. On basis of the experimental results we can say that our solution produces satisfactory results. |
