Smart Drones in Search and Rescue: Leveraging AI for Autonomous Decision-Making

Integrating AI with UAVs (Unmanned Aerial Vehicles) has opened up new opportunities in search and rescue operations. Object detection and recognition in real-time aerial images are crucial for locating missing persons, monitoring natural disasters, and enhancing emergency response efforts. However, challenges remain in accurately identifying and tracking individuals across varying conditions, such as changes in lighting, occlusions, and environmental factors. Our primary goal is to develop a robust UAV-based rescue framework. The proposed solution is aimed to use a Deep Learning (DL) algorithm optimized for seamless integration into extensive computer vision pipelines on edge devices and to be designed to deliver realtime performance in search and rescue missions, enabling rapid and accurate detection of individuals in distress. To represent diverse rescue scenarios we compiled a dataset from publicly available and in-house data. A state-of-the-art detection model is employed to identify individuals including the UAE attire. Realtime processing is achieved by deploying the proposed solution into the embedded device (NVIDIA Jetson Orin). ROS2 establishes a video analytics pipeline capable of processing live drone footage via RTSP streaming. Communication between aerial and ground components is managed via MAVLink protocol. The proposed framework demonstrates strong potential for real-time UAV-based search and rescue operations at the update rate of 30 Hz, enhancing mission responsiveness and effectiveness. With scalability for broader rescue applications, this model aligns with the objectives of the UAE National Emergency Crisis and Disaster Management Authority, supporting rapid response capabilities in various emergency and disaster scenarios.