AI and Drones in Post-Earthquake Search and Rescue

Following a major earthquake, urban search-and-rescue (USAR) teams face a critical race against time to locate survivors trapped beneath rubble. This article explores how modern responders combine high-resolution drone orthomosaics with artificial intelligence (AI) for building-damage segmentation to rapidly assess disaster zones, classify structural destruction, and prioritize rescue operations at the most critical collapsed structures.

Capturing the Disaster Zone with Drone Orthomosaics

Immediately after an earthquake, traditional ground reconnaissance is often blocked by debris and compromised infrastructure. Satellite imagery, while useful, can be delayed by orbital paths, obscured by cloud cover, or lack the resolution needed to see specific structural failures.

USAR teams deploy uncrewed aerial vehicles (drones) to fly automated grid patterns over the affected urban area. These drones capture thousands of overlapping, high-resolution images embedded with GPS data. Photogrammetry software then stitches these individual photos together, correcting for lens distortion, camera tilt, and terrain relief.

The result is an orthomosaic: a single, geographically accurate, high-resolution map of the disaster zone. Unlike a standard aerial photograph, an orthomosaic acts like a true map, allowing teams to take precise measurements of distances, debris fields, and building footprints.

Automated Assessment via Building-Damage Segmentation AI

Analyzing a massive orthomosaic manually takes hours that rescue teams do not have. To accelerate this process, the orthomosaic is fed into a computer vision model trained specifically for building-damage segmentation.

Using deep learning architectures, such as Convolutional Neural Networks (CNNs) or Vision Transformers, the AI scans the imagery at the pixel level. The process typically follows three steps:

  1. Footprint Localization: The AI identifies the boundaries of every structure, often referencing pre-earthquake baseline maps to see what stood there before.
  2. Feature Extraction: The model looks for visual indicators of damage, such as displaced roofs, rubble piles expanding beyond original building footprints, changes in texture, and irregular shadow patterns indicating a sudden loss of building height.
  3. Damage Classification: The AI segments and color-codes each building into specific damage tiers: no damage, minor damage, major damage, or complete collapse.

Prioritizing Collapsed Structures for Rescue Deployment

Once the AI generates the segmented damage map, USAR command centers overlay this data onto Geographic Information Systems (GIS) to make tactical deployment decisions. The prioritization of collapsed structures relies on combining the AI's structural findings with human intelligence data:

  • Targeting Total Collapses: The AI immediately highlights completely collapsed buildings (often referred to as "pancake" collapses), where the highest number of trapped, immobile victims are located.
  • Population and Occupancy Overlay: Teams cross-reference the AI-identified collapses with municipal data. A collapsed high-density residential apartment complex, hospital, or school is prioritized over a collapsed warehouse or commercial building, depending on the time of day the earthquake struck.
  • Structural Vulnerability Filtering: By comparing the AI segmentation with municipal building records, teams can prioritize older unreinforced masonry or non-ductile concrete structures, which are known to cause higher fatality rates when they fail.
  • Route Optimization: Because the orthomosaic shows the entire street grid, teams use the map to identify blocked roads and plot the fastest, safest physical access routes for heavy machinery and K9 units to reach the high-priority structures.

By automating the transition from raw aerial photos to actionable damage maps, drone orthomosaics and segmentation AI compress days of manual reconnaissance into mere hours. This allows USAR teams to bypass intact neighborhoods and immediately direct life-saving resources to the exact coordinates of the worst structural failures.