Human Detection From Unmanned Aerial Vehicles' Images for Search and Rescue Missions: A State-of-the-Art Review

Natural disasters continue to occur at an alarming rate impacting communities worldwide and requiring greater preparedness and response efforts. In the aftermath of disasters such as hurricanes and earthquakes, search and rescue (SAR) usually takes the highest priority on the response list. Utilizing recent advancements in technology, especially Unmanned Aerial Vehicles (UAVs) and Computer Vision in SAR operations has become more common due to their ability to quickly scan large disaster areas, identify and locate victims, and deliver first aid supplies especially in challenging environments. Researchers have leveraged machine learning (ML), particularly Convolutional Neural Networks, to enhance human detection accuracy. However, despite achieving excellent recall rates, the efficiency of these algorithms during actual SAR missions remains a critical consideration. The aim of this paper is to thoroughly examine the literature on human detection from UAV aerial images in SAR scenarios. This paper reviews the existing literature comprehensively and categorizes methods based on disaster type, environment, case study availability, and UAV system capabilities. It also reviews in detail the ML approaches used in the literature and compares them based on factors like image types, training datasets, model details, processing hardware, and evaluation methods. Moreover, it reviews and compares the available datasets according to their types, quantities, applications, and relevance to SAR operations. It also briefly covers state-of-the-art hardware for real-time onboard processing. Finally, we discuss the main challenges and recommendations for achieving fully autonomous UAV systems for human detection in SAR missions and offer recommendations for future research directions.