Research programmes and evidence

Problem

UAV positioning can become unreliable when satellite navigation is blocked, degraded or deliberately disrupted. Visual observations must then be related to georeferenced map evidence despite changes in altitude, season, route and geography.

Why it matters

A dependable visual backup can support safer autonomous flight and mission continuity in GNSS-denied and GNSS-degraded environments.

Approach

• Nadir UAV imagery matched with satellite imagery and orthophotos
• Autoencoder-based representation learning and deep neural networks
• Template matching and cross-view localization
• Terrain-referenced altitude labelling with EXIF and DEM data
• Generalization tests across urban, rural, seasonal and route conditions

Evidence

• The approved 2024 doctoral thesis documents autoencoder and template-matching experiments for terrain-based UAV localization.
• The 2026 peer-reviewed altitude study evaluates a ResNet50 regression model on 303,710 real nadir images.
• Conference records document GNSS-denied localization and vision-based altitude-estimation outputs presented in 2025.

Problem

Image-only analysis can miss the topographic structure that governs scale, surface form, visibility and spatial change. Orthophotos and elevation products must be analysed as a connected geospatial system.

Why it matters

Integrating orthophotos with digital surface, terrain and elevation models supports terrain-aware localization, surface-area measurement, geomorphological change analysis and risk mapping.

Approach

• Orthophoto, DSM, DTM and DEM integration
• SYM and SAM products in Turkish geospatial workflows
• Terrain-referenced visual positioning and altitude labelling
• True surface-area and resolution-sensitivity analysis
• Multi-temporal photogrammetry and GIS-based change mapping

Evidence

• The altitude study combines EXIF GPS altitude with a 30 m DEM to derive above-ground-level labels.
• The public surface-area repository implements DEM-based topographic area experiments and resolution comparisons.
• The Cappadocia project specifies multi-temporal aerial photography, orthophotos, surface models, dense point clouds and GIS integration.

Problem

Archaeological traces can be subtle, spatially extensive and difficult to recognise consistently across large orthophoto and elevation datasets.

Why it matters

A transparent screening workflow can help specialists prioritise candidate areas while preserving georeferencing and the distinction between model candidates and confirmed archaeological evidence.

Approach

• High-resolution orthophoto and elevation-model analysis
• LiDAR-compatible DSM-to-DTM preprocessing
• Multi-band GeoTIFF workflows combining RGB, DSM and DTM
• Deep learning, vision-language models and classical image processing
• GIS vector output for expert review

Evidence

• The institutional project page identifies automatic archaeological site detection from elevation models and high-resolution orthophotos as the research objective.
• The MIT-licensed repository documents a five-band source stack: red, green, blue, DSM and DTM.
• The current software derives SVF, SLRM, slope and normalized DSM channels and supports GeoPackage output.