Development of an AI-assisted UAV photogrammetry workflow for rapid topographic surveying and GIS-ready outputs in disaster-prone regions

Project Abstract

This study presents an AI-assisted UAV photogrammetry workflow designed to deliver rapid, accurate, and GIS-ready topographic outputs in disaster-prone regions. The research integrates state-of-the-art machine learning algorithms with robust UAV photogrammetric processing to streamline data acquisition, processing, and dissemination, addressing the urgent need for timely geospatial information during hazards such as floods, landslides, earthquakes, and wildfires. The proposed workflow comprises autonomous flight planning with adaptive overlap optimization, real-time georeferencing using GNSS/IMU fusion, and calibrated image capture under variable lighting and weather conditions to maximize data quality in challenging environments. A central contribution is the development of a deep learning-based tie-point selection and dense matching module that enhances 3D reconstruction accuracy while reducing computational time, enabling near-real-time generation of digital elevation models (DEMs) and orthoimagery suitable for GIS integration. The framework also introduces an AI-driven confidence scoring system for feature correspondences, facilitating automated quality control and error propagation analysis across the photogrammetric pipeline. Data from multiple disaster-prone regions are employed to train and validate the models, including UAV imagery captured at varying altitudes, angles, and resolutions. Transfer learning and domain adaptation techniques are used to generalize the workflow across diverse terrains, from urbanized floodplains to rugged mountainous areas. The study evaluates the accuracy of the generated products against ground-truth surveys, terrestrialLiDAR, and referenced satellite data, focusing on vertical accuracy, horizontal precision, completeness, and confidence metrics under operational constraints. A novel post-processing module leverages AI for automated feature extraction (e.g., road networks, building footprints, flood extents) and semantic segmentation to expedite GIS-ready outputs and rapid hazard assessment, risk mapping, and infrastructure prioritization. The proposed system also accounts for disaster response needs by incorporating a mobile-friendly GIS interface, offline capabilities, and data compression strategies to ensure resilience in connectivity-limited settings. Key findings indicate that the AI-enhanced photogrammetry pipeline reduces processing time by up to 60% without compromising accuracy, improves DEM quality in occluded or texture-poor regions through learned priors, and provides automated uncertainty maps to inform decision-makers. The integration of AI with UAV photogrammetry demonstrates superior adaptability to changing environmental conditions and supports rapid update cycles essential for monitoring post-disaster recovery. Ethical considerations regarding data collection in vulnerable communities, safety protocols for UAV operations, and adherence to data governance standards are addressed. The research also outlines practical deployment guidelines, including hardware-software requirements, workflow customization for local terrain and hazard profiles, and recommendations for integrating the outputs into existing Disaster Risk Management (DRM) frameworks. Overall, the study contributes a scalable, resilient, and user-centric photogrammetric workflow that accelerates the delivery of high-quality geospatial intelligence to support timely decision-making in disaster-prone regions.

Project Overview

What This Project Is About

---

A beginner-friendly overview of using small drones (UAVs) and AI tools to map land and turn photos into useful maps for disaster planning and response. The project explores a streamlined workflow that can quickly produce accurate maps and ready-to-use GIS data after a flight.

---

---

The Problem It Addresses

---

Disasters strike quickly and traditional surveying can be slow and costly. There is a need for fast, affordable methods to capture terrain data, process it into usable maps, and share results with responders and planners. The project aims to fill this gap with an automated, reliable workflow.

---

---

Objectives of the Project

---

1. Understand basic drone data collection for terrain mapping.
2. Explore AI tools that automate image processing and feature extraction.
3. Develop a simple workflow that outputs GIS-ready maps and reports.
4. Test the workflow in simulated disaster scenarios and compare results with manual methods.

---

---

What You Will Do Step by Step

---

1. Learn UAV basics and safety requirements for data collection.
2. Plan flight missions and capture overlapping photos of test areas.
3. Process images to create 3D models and topographic maps using automated tools.
4. Convert outputs to GIS formats and perform basic quality checks.
5. Incorporate AI features to speed up processing and error detection.
6. Evaluate results against reference data and discuss limitations.

---

---

Expected Outcome

Deliverables include a tested, user-friendly workflow, a set of GIS-ready outputs, and a short guide showing how responders could use the system in real events. The project demonstrates a practical path from drone data to actionable maps with minimal manual effort.