Abstract: Single-object tracking in remote sensing video continuously estimates the spatial state of a specified moving target, thereby capturing its trajectory and state evolution over time and extending remote sensing analysis from static object recognition to dynamic process understanding. Compared with conventional single-object tracking in natural scenes, single-object tracking in remote sensing video faces additional challenges arising from platform motion, long-range imaging, small target size, limited spatial detail, complex background interference, and heterogeneous spectral observations. The methodological development is therefore shaped by observation platform, imaging mechanism, data availability, and onboard computing resource. Single-object tracking in unmanned aerial vehicle videos, satellite videos, and hyperspectral videos is systematically reviewed in terms of observation platforms and data acquisition, training data sources, benchmark datasets, evaluation protocols, and model development. For unmanned aerial vehicle videos, the methodological evolution has broadly followed the development of generic visual tracking, progressing from online discriminative learning and Siamese matching to Transformer-based interaction and one-stream joint encoding. The related methods increasingly emphasize target representation under aerial views, temporal context modeling, interference suppression, and efficient token processing. Single-object tracking in satellite video has gradually moved beyond the direct transfer of generic trackers. Early studies mainly focused on search-region adjustment, template maintenance, motion constraints, and state representation, whereas more recent methods incorporate historical observations into deep matching, preserve high-resolution features for small targets, enhance target-aware interactions, maintain historical states, and explicitly estimate target orientation. Single-object tracking in hyperspectral video seeks to convert spectral responses into discriminative target representations. Representative strategies include explicit material feature construction, band grouping and pseudo-color transformation, joint spatial-spectral modeling, and prompt- or adapter-based transfer from pretrained visual backbones. Several common technical issues arise across these research branches. Small targets and repeated downsampling operations weaken target-specific details, while background clutter and visually similar instances increase the risk of target-distractor confusion. Historical templates, temporal tokens, trajectories, and memory states improve adaptation to appearance and motion changes. However, unreliable predictions can accumulate and propagate over long sequences. Target state modeling therefore needs to jointly organize geometric states, appearance cues, motion trends, visibility, and prediction confidence while controlling error propagation during recursive updates. Multi-source information fusion provides complementary spectral, thermal, material, and semantic cues, but band redundancy, noise, spatial misregistration, missing modalities, and limited training data complicate representation learning and feature interaction. Domain shifts induced by changes in platforms, sensors, illumination, weather, and geographic environments further challenge model generalization. Practical deployment requires joint consideration of tracking accuracy, computational latency, memory consumption, parameter adaptation costs, and robustness to changing inputs. The existing research has shifted from the direct transfer of generic visual models toward specialized designs tailored to different observation platforms, imaging mechanisms, and continuous-observation characteristics. Nevertheless, the dedicated training resources remain limited, long-term target state maintenance remains vulnerable to error accumulation, and multi-source input adaptation has yet to be coordinated with efficient operation on real remote sensing platforms. Future research should strengthen remote sensing video data and training foundations, advance sequence-level target state modeling, and coordinate multi-source input adaptation with resource-aware edge deployment. These advances will help transform continuous remote sensing observations into reliable target trajectories and state information for traffic monitoring, maritime surveillance, emergency response, and dynamic scene understanding.