A Comprehensive Review of Deep Learning for Disaster Victim Detection: Trends, Challenges, and Future Directions
Harshitha Gudavalli, Ramachandra Rao Kurada, Srikanth Pala, Ramu Yadavalli, Sunil Pattem, Kiran Sree Pokkuluri · 2025
The swift and precise identification of victims in disaster situations is essential for improving emergency response efforts and reducing casualties. Traditional search- and-rescue methods often face inefficiencies caused by environmental challenges, occlusions, and the necessity for manual intervention. Recent advancements in deep learning-based object detection models, particularly Convolutional Neural Networks (CNNs), YOLO, Faster R-CNN, SSD, and transformer-based architectures, have significantly enhanced the speed and accuracy of victim identification. This review provides a thorough analysis of various deep learning techniques used for disaster victim detection. It compares their effectiveness and explores the latest trends, such as multimodal data fusion, UAV-based real-time searches, and self-supervised learning. The paper also discusses significant challenges, including how to handle occlusions, constraints associated with real-time deployment, and limitations related to available datasets. Additionally, the review emphasizes future research directions, such as the incorporation of thermal imaging, LiDAR-based sensing, the deployment of edge AI, and transformer-based object detection models. This review seeks to guide future research toward more efficient and scalable AI-driven disaster response systems by summarizing advancements, challenges, and potential innovations.