🌍 Automated Disaster Damage Assessment from Aerial Imagery

This project was developed as part of our Senior Capstone at The University of Texas at Dallas, where we apply academic research and engineering principles to solve real-world problems.

Under faculty guidance, our team designed and implemented an AI-powered system that translates recent advances in Vision-Language Models (VLMs) into practical disaster response tools.

---

📌 Project Background

Natural disasters can rapidly devastate entire communities, making fast and accurate damage assessment critical for response and recovery efforts.

Traditional assessment methods rely heavily on manual inspection of aerial imagery, which is:

• Time-consuming
• Costly
• Difficult to scale

This project leverages recent advances in Vision-Language Models (VLMs) to automate damage analysis using pre- and post-disaster imagery, enabling quicker insights and more informed decision-making.

---

🎯 Project Objectives

1️⃣ Vision-Language Model Pipeline

• Develop a VLM-based pipeline to analyze pre- and post-disaster aerial imagery.
• Automatically generate structured damage assessments.
• Note: Fine-tuning or training a new model is not required.

---

2️⃣ Interactive Geospatial Dashboard

• Design and implement a web-based dashboard that:
  • Displays aerial imagery
  • Overlays model-generated damage classifications
  • Enables intuitive geospatial interaction

---

3️⃣ Conversational Chatbot Interface

• Build a chatbot that allows users to:
  • Query disaster impacts
  • Retrieve information about specific addresses or streets
  • Interact with model outputs in natural language

---

4️⃣ Evaluation Module

• Implement an evaluation system that:
  • Compares model predictions against FEMA ground-truth labels
  • Measures performance and accuracy

---

5️⃣ Documentation & Reporting

• Maintain thorough documentation covering:
  • Development progress
  • Design decisions
  • Testing results
  • Milestone completion

---

🚀 Expected Impact

By combining computer vision, natural language processing, and geospatial visualization, this project aims to:

• Improve disaster response timelines
• Reduce reliance on manual image inspection
• Support data-driven recovery decisions
• Demonstrate practical applications of academic AI research

👥 Team Members

Senior Capstone Team — University of Texas at Dallas

• Kanchan Javalkar
• Shraddha Subash
• Nadeeba Atiqui
• Noel Emmanuel
• Anjali Kadur
• Adarsh Goura

Faculty Advisor: Dr. Semih Dinc