Dataset preparation for swimmer detection

The large amount of data that is created every day can be used to develop artificial intelligence algorithms in the domain of computer vision that solve tasks such as image classification, face detection and action recognition. These datasets are most often created from videos and images downloaded from television channels or the YouTube social network and are collected and prepared for the appropriate task. We were interested in the task of detecting swimmers, so that the model could be used to recognize and improve swimming techniques. Although today there are huge open image databases like COCO and ImageNet, prepared for supervised machine learning and sports scene databases like Olympic Sports Dataset, UCF Action Sport dataset or Sport-1M that include images of more popular (watched) sports, none of them include images that could be used to make our swimmer detection model. Therefore, this paper describes the process of recording and collecting video material and preparing a set of UNIRI-SWM images for swimmer detection. The set includes shots of swimmers in real, situational training and competition conditions filmed by action cameras from different shooting angles. The paper presents the results of swimmer detection using deep convolutional neural networks Mask R-CNN and Yolo v3, learned in the set of general images before and after learning in the set UNIRI-SWM. The results show that after adjusting the model on the appropriate set of images from the swimming domain, very good results of swimmer detection can be achieved.