A comprehensive survey on unmanned aerial vehicle-based single object tracking

With the widespread adoption of unmanned aerial vehicle (UAV) technology in security surveillance, disaster response, agricultural monitoring, and beyond, single-target tracking has become a cornerstone for enabling autonomous UAV operations. This paper provides a comprehensive review of core technologies and recent advancements in UAV-based single object tracking(SOT), covering both traditional filtering methods (e.g., Kalman filter, particle filter, mean shift, and their variants) and deep learning-driven end-to-end frameworks, including Siamese networks and Transformer-based architectures. We begin by outlining the technological evolution of UAVs and the practical significance of single-target tracking in real-world applications. Subsequently, we critically analyze technical bottlenecks encountered in scenarios involving target occlusion, scale variation, dynamic environmental interference, and onboard hardware limitations. To address these challenges, this work explores potential solutions such as multi-modal data fusion, online model adaptation, and lightweight model architectures tailored for resource-constrained UAV platforms. Furthermore, we systematically summarize widely adopted benchmarks (e.g., VisDrone, UAV123) and evaluation metrics, emphasizing their applicability in UAV-specific contexts. Through empirical case studies and comparative analyses, we demonstrate the performance discrepancies among different algorithms in practical UAV deployments. Finally, we propose future research directions focused on long-term tracking robustness, cross-domain generalization, and edge computing optimization, aiming to bridge the gap between theoretical advancements and real-world implementation.