Automation of Athletes Identification in an Image

This paper deals with the task of athlete recognition in images from sports competitions. Identification of an athlete by recognising his start number allows to avoid problems connected with recognising athletes by face and allows to automate the process of tagging images from competitions. The paper provides a brief overview of various existing methods for localising objects and recognising text in images. The authors prepared and labelled several image datasets and described them in the paper. The authors developed models, which are based on the neural network architecture YOLO, to localise the athletes, filter out images that are unsuitable for number recognition, and recognise digits in athlete start numbers with high accuracy. Based on a combination of these models, the authors developed the start number recognition model, whose testing metrics are given in the paper.