Single object tracking（SOT） is recognized as one of the fundamental tasks in the field of computer vision. However, traditional correlation filters and Siamese network architectures struggle to meet the growing demands for accuracy and robustness in complex and dynamic environments. Transformer exhibits significant advantages in SOT by virtue of its powerful global modeling capability. Therefore, the recent research advances in Transformer-based SOT are reviewed systematically. Based on the overall pipeline design, existing tracking algorithms can be categorized into two primary types： two-stream two-stage algorithms and one-stream one-stage algorithms. Representative algorithms of each category are analyzed in depth to highlight their relations and characteristics, while the research status of lightweight Transformer-based tracking methods is summarized. In addition, recent emerging trends, such as Mamba-based tracking algorithms and unified model architectures, are further investigated and their promising potential in model efficiency and generalizability is discussed as well. The performance of different Transformer-based tracking methods is comprehensively analyzed and evaluated on multiple mainstream datasets. Finally, several promising directions for future research in SOT, including lightweight models, multimodal fusion, long-term tracking and foundation models-driven tracking, are outlined, providing valuable references for the research and development of SOT.