부산대학교 도서관

Tiny objects are often found in modern practical detection and tracking tasks, they have a weak appearance and characteristics, which leads to the need for the formation of special tracking methods. The main datasets for evaluating and training algorithms contain a small number of sequences with tiny objects, so methods for such tasks need to be evaluated on additionally generated dataset. This article focuses on the algorithm for tracking small-sized moving objects. A dataset based on a sample of video sequences with tiny objects from common datasets is also proposed for evaluation. The proposed tracking method is compared with other popular algorithms on this dataset.