Real-Time Edge Intelligence in UAV for Search-and-Rescue: Onboard Energy-Efficient Video Summarisation with Reduced Data Transmission

Rapid detection of individuals in search and rescue (SAR) operations is critical for minimising casualties and enabling effective relief coordination. Unmanned aerial vehicles (UAVs) provide real-time situational awareness, but continuous high-resolution video transmission causes bandwidth constraints, delays decisions, and drains onboard energy. This paper introduces a modular UAV system for energy-efficient onboard video summarisation tailored for SAR. The system integrates lightweight YOLOv5n object detection and histogram-based keyframe extraction on a Raspberry Pi 4, transmitting only critical frames to ground teams. Built with off-the-shelf components, the UAV optimises weight and power while supporting on-device edge processing. The results show significant reductions in energy consumption and data transmission compared to full-video processing, maintaining high detection accuracy. This adaptable, low-power framework enhances mission duration and decision-making speed, offering a scalable solution for bandwidth constrained, energy-aware aerial monitoring in disaster response scenarios.