Enhancing Earthquake Response With AI: Real-Time Human Detection Using Deep Learning and Robotic Systems

This paper presents a practical AI-enabled Earthquake Rescue System (ERS), which applies a deep learning technology-based computer vision to identify the trapped human victims under the rubble caused by an earthquake. In the traditional rescue operation, the physical characteristics of a voice, motion, and temperature are employed to locate the trapped victims. However, the effectiveness of the rescue operation is low because the traditional characteristics are not effective in earthquake rescue scenarios. To overcome the limitations of the traditional rescue operation, this work proposes an AI-enabled rescue operation based on the application of a YOLOv8 model to identify the trapped victims under the rubble caused by an earthquake. The proposed model applies a dataset of 6,837 images to fine-tune a pretrained YOLOv8 model through transfer learning. The experimental results show that the proposed model has achieved a mean average precision of 93.1%, recall of 90.5%, F1-score of 90%, precision of 98.2% in face detection, 94.6% in palm detection, and 88.7% in the detection of other body parts. The fine-tuned model is then implemented on a low-cost NVIDIA Jetson Nano, which is integrated with a mobile robotic vehicle and a camera. The system facilitates near real-time analysis and cloud-based alert transmission to aid the rescue teams in quickly locating the possible location of survivors. The results obtained demonstrate the possibility of implementing lightweight AI-assisted rescue systems to improve the efficiency of earthquake rescue operations.