Disaster Response AI: Real-Time Victim Localization from Images and Videos in Natural Calamities

Abstract

Natural disasters such as earthquakes and floods cause severe destruction and lead to significant loss of human lives. Rapid identification and localization of victims are crucial for effective rescue operations during such emergencies. However, traditional search and rescue methods are often slow and require extensive manpower, which can delay assistance to those in need. Recent advancements in Artificial Intelligence (AI) provide new opportunities to enhance disaster response systems. By analyzing real-time images captured through drones and surveillance cameras, AI models can quickly detect and locate victims in affected areas. This project proposes an AI-based system designed to automatically detect and localize victims using aerial imagery from disaster zones. The proposed approach aims to improve the speed and accuracy of victim identification, thereby assisting rescue teams in making faster decisions and enhancing the overall efficiency of disaster response operations.

Introduction

Natural disasters—earthquakes, floods, landslides, hurricanes—pose major challenges for search and rescue operations. Manual victim search is time-consuming, labor-intensive, and risky, often delaying rescue and lowering survival rates. Disaster areas may include collapsed structures, flooded zones, and blocked roads, limiting human access.

Problem:
Traditional search methods struggle with large-scale, hazardous environments. Real-time identification and localization of victims is critical to improving emergency response outcomes.

Proposed Solution:
The system leverages Artificial Intelligence (AI), deep learning, and computer vision to automate victim detection from aerial or surveillance images:

1. Data Acquisition:

   • Images captured via drones or cameras, including simulated disaster scenarios with debris and damaged structures.

   • Publicly available human image datasets supplement real-world scarcity.

2. Preprocessing:

   • Noise removal using median filtering and data cleaning to enhance image quality.

3. Victim Detection:

   • Uses Convolutional Neural Networks (CNNs) and real-time object detection algorithms (e.g., YOLOv5) to identify humans in complex environments.

   • The model predicts bounding boxes with confidence scores indicating likely victim locations.

4. System Workflow:

   • Input: Real-time images from drones/cameras

   • Processing: CNN-based detection

   • Output: Detected victim locations with bounding boxes, assisting rescue teams in prioritizing search areas

Results:

• The proposed model was tested against HOG+SVM, Faster R-CNN, and YOLO-based detection methods:

• The system outperforms traditional and existing deep learning approaches in accuracy and enables real-time victim detection, supporting faster and more effective disaster response.

Conclusion

This study presented an AI-based victim detection system designed to assist rescue teams in disaster-affected areas. The proposed system utilizes aerial images and deep learning–based image analysis techniques to automatically detect human presence in disaster environments. By integrating image acquisition, preprocessing, and detection models, the system improves the efficiency of victim identification compared to traditional manual search methods. The experimental results demonstrate that the proposed approach achieves better detection accuracy and faster processing, enabling quicker identification of victims in large and complex disaster zones. This can significantly support rescue teams in making timely decisions and improving the effectiveness of disaster response operations. Future work will focus on enhancing the system by incorporating more advanced deep learning models, expanding disaster image datasets, and integrating real-time drone-based monitoring to further improve detection performance and support efficient disaster management.

References

[1] Krizhevsky, I. Sutskever, and G. E. Hinton, “ImageNet classification with deep convolutional neural networks,” Advances in Neural Information Processing Systems, vol. 25, pp. 1097–1105, 2012. [2] [2] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, “You Only Look Once: Unified, real-time object detection,” in Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), 2016. [3] [3] S. Ren, K. He, R. Girshick, and J. Sun, “Faster R-CNN: Towards real-time object detection with region proposal networks,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015. [4] [4] M. Erdelj and E. Natalizio, “UAV-assisted disaster management: Applications and open issues,” in Proc. IEEE Int. Conf. Computing, Networking and Communications, 2016. [5] [5] A. Bochkovskiy, C. Y. Wang, and H. Y. M. Liao, “YOLOv4: Optimal speed and accuracy of object detection,” arXiv preprint arXiv:2004.10934, 2020. [6] [6] R. Gupta et al., “Creating xBD: A dataset for assessing building damage from satellite imagery,” in Proc. IEEE/CVF Conf. Computer Vision and Pattern Recognition Workshops, 2019. [7] [7] C. Kyrkou and T. Theocharides, “EmergencyNet: Efficient aerial image classification for drone-based emergency monitoring using deep learning,” IEEE Transactions on Emerging Topics in Computing, 2021. [8] [8] Y. Zhang et al., “Victim detection from UAV images using deep learning in disaster scenarios,” Remote Sensing, vol. 14, no. 13, 2022.

Copyright

Copyright © 2026 Aryan Jagaty, Mrunal Futane, Raj Gupta, Pradeep Jha . This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.