UAV Disaster Mapping

Project Overview

This ongoing research exploration investigates low-cost unmanned aerial vehicles (UAVs) equipped with onboard edge AI to rapidly map landslide corridors, collapsed infrastructure, and evacuation routes in mountainous regions like Nepal following natural disasters.

The Challenge & Problem

Following earthquakes and landslides in Nepal, steep terrain and severed roadways make ground reconnaissance impossible, while commercial satellite imagery suffers from cloud cover and update delays.

The Engineering Solution

An integrated framework combining autonomous drone flight paths, edge AI semantic segmentation for debris detection, and rapid orthomosaic generation on portable field computing stations.

System Architecture

• Autonomous survey flight grid generator factoring rugged mountain topography
• YOLOv8-based edge damage detection model trained on aerial rubble imagery
• OpenSfM photogrammetry pipeline for rapid digital surface model (DSM) creation
• Offline GIS viewer for emergency first responders

Key Capabilities & Features

• Automated road blockage detection and passability classification
• 2D orthomosaic and 3D elevation point cloud generation from drone stills
• Onboard inference on lightweight edge compute (Jetson Orin Nano / Raspberry Pi)
• Low-bandwidth mesh network telemetry relay