LEGG AI: Ohio State’s New Tool Reconstructs Earthquake Damage in 3D
- What: The LoRA-Enhanced Ground-view Generation (LEGG) diffusion model.
- Developer: Researchers at The Ohio State University.
- Function: Generates photorealistic 3D ground-level reconstructions from aerial drone imagery.
- Case Study: Successfully tested on data from the 7.8 magnitude 2023 Kahramanmaraş, Turkey earthquake.
- Goal: To accelerate emergency response times and improve infrastructure resilience.
Researchers at The Ohio State University have developed a new "imaginative" AI tool that can reconstruct photorealistic 3D ground-level views of earthquake damage using only aerial drone footage. The LoRA-Enhanced Ground-view Generation (LEGG) diffusion model fills a critical gap for first responders, who often lack ground-level visibility in the immediate aftermath of a disaster when street access is blocked or dangerous.
By bridging the divide between aerial data and street-level reality, the LEGG model allows emergency crews to identify structural failures that are invisible from the air. According to a study published in the International Journal of Remote Sensing, the AI can recognize complex visual patterns, such as building tilts and façade cracks, which are vital for making split-second rescue decisions.
Bridging the Aerial-to-Ground Gap
Traditional damage assessment relies heavily on Unmanned Aerial Vehicles (UAVs) or lidar-based detection to survey collapsed buildings from above. However, these methods often fail to capture the perspective needed by rescuers on the ground. Manual assessments to fill these gaps can take days or weeks—a timeline that is often too slow for rapid recovery missions.
"What our algorithm does is generate thousands of pairs of semi-realistic photos of what a building looks like on the top and from the ground," said Rongjun Qin, co-author of the study and a professor of civil, environmental, and geodetic engineering at Ohio State. "Having such data is vital, as drones gather important information from above, but people actually make emergency decisions from ground-level views."
Tested on the Kahramanmaraş Disaster
To validate the model, the research team applied LEGG to the 2023 Kahramanmaraş, Turkey, earthquake. This 7.8 magnitude event destroyed approximately 280,000 buildings and damaged at least 700,000 others. The AI was trained on a dataset of just 3,000 city structures, comparing drone imagery from 2015 to photos taken shortly after the 2023 quake.
Despite the limited training set, the model successfully extracted subtle cues to generate high-resolution street-level views. It accurately identified partial collapses and structural tilts that would have been difficult to confirm through top-down imagery alone. This "imaginative" capability stems from the AI's ability to synthesize what a structure should look like based on the aerial patterns it recognizes.
Impact on Disaster Management
The LEGG model represents a shift in how engineers and government agencies can approach disaster preparedness and response. By creating 3D simulations of hypothetical earthquakes, the tool could help urban planners design more resilient infrastructures.
"This work presents a great opportunity for engineers and other decision makers to remotely assess the damage in structures soon after a disaster," said Halil Sezen, co-author and professor of structural engineering at Ohio State.
For developers and the AI industry, this project highlights the power of Low-Rank Adaptation (LoRA) and diffusion models in high-stakes, real-world applications. It demonstrates that AI can do more than generate art; it can serve as a predictive engine for civil safety.
"This AI bridges the gap between what a drone sees and what a rescuer needs, turning aerial data into actionable ground-level maps when every second counts."
What’s Next
The researchers noted that the LEGG algorithm is intended to work in tandem with existing emergency and resource planning tools. Future experiments will aim to test the model in other earthquake-prone regions, specifically mentioning Japan and California.
As the quality and quantity of drone data improve, the team expects the AI to become a "generous predictor" of both past damage and future outcomes, potentially reshaping how global emergency management policies are formed.
Sources
All technical specifications, pricing, and benchmark data in this article are sourced directly from official announcements. Competitor comparisons use publicly available data at time of publication. We update our coverage as new information becomes available.