Vision-Radar Fusion for UAV Perception in Low-Altitude Emergency Rescue: A Survey

Abstract

Unmanned Aerial Vehicles (UAVs) have emerged as critical tools in emergency rescue operations, particularly in urban low-altitude complex environments characterized by dynamic obstacles, adverse weather, and signal interference. However, traditional single-modality perception systems, reliant on either vision or radar, struggle with limitations such as motion blur, low light conditions, and occlusion, compromising detection accuracy and navigation safety. This survey provides a comprehensive review of vision-radar fusion perception systems tailored for UAVs in such scenarios. We first examine advancements in image quality enhancement techniques, including diffusion models and super-resolution methods like SRGAN  and ESRGAN , which address degradation in low-altitude imagery. Next, we explore multi-target detection and recognition via multi-modal fusion frameworks, such as Faster R-CNN-based approaches  and cross-view spatial fusion. We then discuss robust multi-target tracking and trajectory prediction, highlighting graph neural networks (GNNs) and Transformer-based models like TrackFormer and MOTR . Finally, we cover low-altitude environment assessment and autonomous path planning, leveraging graph theory, 3D risk mapping, and deep reinforcement learning (e.g., MADDPG). By synthesizing global research trends, challenges like spatio-temporal alignment and real-time processing, and future directions toward end-to-end multi-modal integration, this review underscores the potential of fusion systems to enhance UAV resilience and efficiency in rescue missions. Our analysis reveals a clear trajectory toward AI-driven, adaptive perception, with implications for urban safety and disaster response.