Automated UAV-based Photogrammetric 3D Mapping and Change Detection for Urban Redevelopment Monitoring using Deep Learning Segmentation

Project Abstract

This study presents an integrated framework for automated UAV-based photogrammetric 3D mapping and change detection to support urban redevelopment monitoring, leveraging deep learning segmentation to enhance feature extraction, classification, and temporal analysis. The workflow combines high-resolution UAV imagery with Structure-from-Motion (SfM) and Multi-View Stereo (MVS) to generate dense 3D reconstructions and accurate orthoimagery across multiple time epochs. A novel end-to-end pipeline is developed to automate data acquisition planning, terrestrial and aerial data fusion, and georeferencing using ground control points and GNSS-enabled UAV flights, ensuring centimeter-level accuracy suitable for urban cadastral and planning applications. The photogrammetric process yields multi-scale 3D city models, digital surface models (DSMs), digital terrain models (DTMs), and semantic-rich orthoimages, which are subsequently integrated into a deep learning segmentation framework. Convolutional neural networks (CNNs) and transformer-based architectures are trained on labeled multi-temporal datasets to classify urban features such as buildings, roads, green spaces, water bodies, construction zones, and demolition sites. The segmentation models are designed to be robust to occlusions, varying illumination, seasonal changes, and sensor inconsistencies, enabling reliable extraction of construction boundaries, material changes, and land-use transitions. Change detection is performed through a hybrid approach that fuses pixel-wise semantic differences with geometry-aware metrics derived from the 3D models, including surface height alterations, volumetric change estimation, and roof/ façade modifications, thereby reducing false positives common in purely 2D analyses. The research investigates automated change detection workflows that can be deployed in near-real-time for monitoring ambitious urban redevelopment projects. An emphasis is placed on scalability, computational efficiency, and open data interoperability, with components implemented in modular microservices and containerized environments. The framework includes quality assurance protocols, uncertainty quantification, and error propagation analysis to provide confidence intervals for mapped changes and 3D metrics. A synthetic-to-real transfer strategy is employed to augment limited labeled data, using generative models to create realistic urban scenes and augment the training set for segmentation and change detection. The effectiveness of the proposed system is validated across diverse test sites characterized by complex morphology, dense built environments, and dynamic land-use evolution. Evaluation metrics include accuracy, precision, recall, F1-scores for segmentation, as well as Chamfer distance, root mean square error (RMSE) for 3D reconstruction, and IoU-based change detection scores. The results demonstrate improved consistency and comparability of urban redevelopment metrics over time, enabling planners and stakeholders to quantify progress, detect illegal or unauthorised construction, assess environmental impact, and inform policy decisions. The study also discusses limitations such as occlusion-driven inaccuracies in dense urban canopies, computational demands for large-scale projects, and the need for standardized protocols to ensure cross-site interoperability. Recommendations are provided for integrating the framework into existing urban planning information systems and open data initiatives to foster transparent, data-driven redevelopment strategies.

Project Overview

What This Project Is About

---

A plain-language overview of using drones to create 3D maps of cities and automatically detect changes over time with simple computer-based image analysis. The project combines aerial photos, 3D reconstruction, and lightweight software to highlight redeveloped areas or new constructions. It focuses on making urban monitoring faster, cheaper, and more accurate for planners and researchers.

---

---

The Problem It Addresses

---

Cities change quickly, but keeping up with accurate maps and change records is hard. Traditional methods are labor-intensive and may miss small or informal developments. This project aims to provide a repeatable, scalable method to track urban redevelopment, helping decision-makers plan infrastructure, zoning, and risks.

---

---

Objectives of the Project

---

1. Collect high-quality drone imagery over selected urban areas. 2. Build a 3D model of the area using photogrammetry. 3. Develop a simple change-detection workflow to compare maps over time. 4. Apply a light deep-learning model to segment buildings, roads, and green spaces. 5. Validate results against ground-truth data and assess accuracy. 6. Create an easy-to-use visualization tool for stakeholders.

---

---

What You Will Do Step by Step

---

1. Plan data collection: choose sites and flights for consistent imagery. 2. Acquire UAV photos and preprocess them (calibration, georeferencing). 3. Generate a 3D reconstruction and textured model from images. 4. Prepare a baseline map and a later map for a redevelopment period. 5. Run a change-detection method to identify differences. 6. Train or adapt a simple segmentation model to classify features. 7. Validate outputs with field data and existing maps. 8. Build a user-friendly interface to view results.

---

---

Expected Outcome

An integrated workflow for 3D urban mapping and change detection using UAV data and basic segmentation, plus a demonstrable case study and a visualization tool. The project should produce accurate change maps, improved understanding of redevelopment patterns, and a practical approach that can be adopted by city planners and researchers.