Single target tracking in high-resolution satellite videos: a comprehensive review

Abstract

ABSTRACT With the development of dynamic acquisition technology for satellite video data, significant progress in target tracking has been made in recent years. This progress plays an important role in monitoring rapidly changing events. Different from targets in ordinary videos, targets in satellite videos usually demonstrate the phenomenon of a small size occupation pattern (small) and weak feature capture (dim) due to occlusion, illumination variation, and confusion with the surroundings. To the best of our knowledge, while some review work has been proposed for satellite videos object tracking, it has primarily concentrated on new datasets or has simply applied target tracking algorithms designed for ordinary videos directly to satellite videos for review. There is a lack of a comprehensive overview of satellite videos data sources and previous satellite video target tracking algorithms. Hence, there is a need for a systematic summary and in-depth exploration of the theory and the present state of satellite video object tracking. We provide an overview as well as a meta-analysis of satellite video targets, data sources, performance metrics, state-of-art models, and the trend of the target-specific comparative methods. To identify emerging research trends and opportunities, we classify established and emerging algorithms into three main categories and evaluate their performance on typical satellite videos. By providing comprehensive comparisons from both experimental (qualitative and quantitative) and theoretical perspectives, we demonstrate the tracking effects of several popular methods in various situations to show their strengths and weaknesses. In addition, this paper also collects the latest open-source datasets and predicts promising future research directions. Our collections of satellite videos benchmark datasets are available at: https://zenodo.org/record/8425369.