Multimodal Data Fusion for UAV Detection and Tracking: A Systematic Review

The fusion of multimodal data has been investigated to increase the robustness of detection and tracking systems in Unmanned Aerial Vehicles (UAVs), especially in adverse scenarios. This paper presents a Systematic Literature Review (RSL) on Deep Learning-based multimodal fusion methods applied to this domain, conducted according to the PRISMA methodology. The search covered publications between 2021 and 2025 in four scientific databases, resulting in the selection of 81 primary studies. The papers were analyzed for evaluation metrics, datasets, and architectural families, including Convolutional Neural Networks (CNN), YOLO, Transformers, and Mamba. The results indicate evolution from convolution-based methods to architectures with attention, dynamic fusion between modalities, and long-range dependency modeling. The review highlights the importance of balancing computational accuracy and cost, in addition to pointing out gaps related to experimental standardization and challenging multimodal datasets.