Architectural Evolution of UAV Tracking Under Efficiency Constraints

UAV tracking is important for aerial surveillance, inspection, and autonomous perception, yet its progress is constrained by the tension between tracking robustness and limited onboard computation. Compared with existing UAV tracking surveys, this review examines UAV tracking from the perspective of architectural evolution under efficiency constraints, and incorporates Mamba- and SSM-based trackers into the analysis. Specifically, this review discusses UAV tracking as a deployment-constrained problem, analyzes CF, Siamese/CNN, Transformer, and Mamba/SSM trackers from a cross-paradigm perspective, and explains how the literature-reported benchmark results should be interpreted under heterogeneous evaluation settings. We then examine how these architectural paradigms, including recent state-space and Mamba-style models, balance representation ability, interaction strength, temporal modeling, and deployment cost under UAV tracking constraints. Finally, we summarize architecture-level trade-offs and outline open problems in preserving local details during sequence modeling, reproducible efficiency evaluation, hardware-aware design, and multimodal UAV tracking.