Application of Convolutional Networks to Detection of Playing Cards

U radu je prikazana primjena konvolucijskih mreža u detekciji igraćih karata. Prije same detekcije generiran je umjetni skup podataka na kojem modeli detekcije uče svoje težine. Objašnjeni su različiti pristupi detekciji objekata te metode mjerenja uspješ- nosti detektora. Implementirane su različite metode detekcije igraćih karata te je ispi- tana njihova točnost na ručno izrad̄enom skupu fotografiranih slika. Dodatno se za sve metode mjeri prosječno vrijeme potrebno za detekciju karata na slikama.The paper presents the application of convolutional networks in the playing card detection problem. Before object detection itself, the dataset on which the detection model trains its weights was generated. Different approaches to object detection and methods for measuring detector performance are explained. Different methods of de- tecting playing cards have been implemented, and their accuracy has been calculated on a manually created set of photographed images. In addition, the average time requ- ired to detect objects in images for all methods was measured