Abstract: This paper aims to review the research progress in remote sensing satellite video single-target tracking technologies, analyze the advantages and disadvantages of existing methods, and explore future development directions. Through literature review and comparative analysis, the research achievements in this field over the past five years were systematically summarized. The existing methods are categorized into two types： correlation filtering-based methods and deep learning-based methods. The technical features and performance of each category were analyzed. Tracking accuracy of representative methods was evaluated based on publicly available datasets, and the applicability and limitations of different methods were discussed. Experimental results show that correlation filtering-based methods perform excellently in terms of computation speed and tracking accuracy. On the publicly available SatSOT dataset, the highest tracking accuracy can reach 69.8%, with an average frame rate exceeding 30frame/s, demonstrating strong practicality and real-time performance. These methods efficiently track targets with low computationalcost by utilizing appearance features and motion information, making them particularly suitable for resource-constrained onboard platforms. In contrast, deep learning-based methods have significant advantages in feature representation and adaptability to complex scenes, but due to the lack of large-scale annotated data in the remote sensing domain, their highest tracking accuracy on the SatSOT dataset is currently 66.9%, slightly lower than correlation filtering methods. This paper summarizes the research progress in remote sensing satellite video single-target tracking. Correlation filtering methods are mature and highly real-time, suitable for current tasks. Deep learning methods show great potential but require further improvements in model optimization.

Key words: satellite videos, single object tracking, correlation filter, deep learning