Detection and tracking of players in sport video

U ovom radu razmatramo problem detekcije i praćenja nogometaša u video snimkama. Ovaj problem sadržava male objekte koji se brzo kreću, međusobno preklapaju i variraju u veličini. U radu su objašnjene konvolucijske neuronske mreže, algoritmi detekcije i algoritmi praćenja. Implementiran je sustav za detekciju igrača temeljen na algoritmima dubokog učenja YOLOv5 i YOLOv8, a praćenje je ostvareno SORT i deepSORT algoritmima. Rezultati sustava se vizualiziraju i vrednuju računanjem kvantitativnih MOT16 metrika. U radu su na petominutnom videu ostvareni zadovoljavajući rezultati MOTA mjere od 65%.In this paper, we consider the problem of detecting and tracking football players in sports video. This problem involves small objects that move quickly, overlap each other, and vary in size. Convolutional neural networks, detection algorithms and tracking algorithms are explained in the paper. A player detection system based on the YOLOv5 and YOLOv8 deep learning algorithms was implemented, and tracking was achieved using SORT and deepSORT algorithms. The results of the system are visualized and evaluated by calculating quantitative MOT16 metrics. The results of the MOTA metric of 65% were achieved on a five-minute video