Oblique aerial imagery for NMA : some best practices

F. Remondino, I. Toschi, M. Gerke, F.c. Nex, D. Holland, A. McGill, J. Talaya Lopez, A. MagarinosISPRS : Proceedings of the XXIII ISPRS Congress : From Human History to the Future With Spatial Information, 12-19 July , Prague, Czech Republic. Peer Reviewed Annals, Volume III-2, . Comm II, ThS14 Recent Developments in Open Data 2016 · OpenReview

Abstract

As urban environments demand increasingly detailed 3D models, National Mapping Agencies (NMAs) are shifting from traditional vertical photography to oblique aerial imagery. This paper synthesizes best practices and R&D experiences from across Europe, highlighting how "Maltese-cross" camera configurations are revolutionizing 3D city modeling while introducing significant challenges in data volume and processing complexity.

Introduction

For decades, aerial mapping relied almost exclusively on nadir (vertical) imagery. While excellent for 2D maps, nadir views struggle with the "vertical dimension" of modern cities: they cannot see building facades, they miss features under overhangs, and they often fail to capture narrow streets in dense urban canyons.

As the "Smart City" concept gains traction—requiring precise 3D models for noise modeling, solar potential analysis, and urban planning—traditional pipelines are hitting a wall. The challenge addressed here is how to transition NMAs from 2D-centric workflows to oblique-based 3D workflows, balancing the superior information density of slanted views with the massive computational and operational costs they incur.

Methods and Technical Approaches

The paper discusses the use of multi-camera systems, typically arranged in a "Maltese-cross" configuration: one nadir camera and four slanted (oblique) cameras pointing in cardinal directions.

Figure 4: Multi-view oblique imagery of a church, demonstrating the different perspectives captured by a Maltese-cross camera configuration.

Key technical approaches include:

  • Automated Aerial Triangulation (AT): Developing strategies to align oblique images, which is significantly harder than nadir AT due to extreme scale variations and radiometric changes between views.
  • Dense Image Matching (DIM): Using multi-view pixel-based algorithms to generate high-density 3D point clouds.
  • LOD2 Modeling: Automatically deriving "Level of Detail 2" building models (which include roof structures) by combining nadir and oblique datasets.
  • Software Integration: Testing commercial and in-house tools like Tridicon/Hexagon, Acute3D, and UltraMap to see how they handle the increased data load.

Key Insights and Results

Oblique imagery isn't just for 3D; it can simultaneously provide 3D point clouds, Digital Surface Models (DSM), true orthoimages, and textured facades, potentially replacing multiple specialized flights. Furthermore, oblique views are far more intuitive for non-expert users (e.g., city planners or citizens) than vertical views, as they mimic how humans naturally perceive the world.

Figure 3: Comparison between (a) a full textured 3D mesh and (b) extracted LOD2 building models derived from oblique aerial datasets.

However, the transition involves significant trade-offs:

  • Data Volume Explosion: For a 25 sq. km area, a traditional nadir flight requires ~400 images and <1 hour of flight time. An oblique flight for the same area requires >6,000 images and >3 hours of flight time.
  • Resolution Realities: ICGC Spain found that due to the Bayer filters used in medium-format oblique cameras, the "effective" resolution is often 1.32x to 2.21x lower than the nominal Ground Sample Distance (GSD).
  • Modeling Success: In the UK study, the City Modeller tool showed higher success rates in sub-urban areas with uniform building shapes compared to complex city centers, where unique architectures still require manual editing.

Limitations and Future Work

The transition to oblique imagery is not yet "plug-and-play." The authors note several hurdles:

  • Processing Bottlenecks: Current off-the-shelf software often fails at the Aerial Triangulation stage for very large oblique blocks, requiring manual intervention to subdivide projects.
  • Illumination Issues: Capturing five views simultaneously means some cameras will inevitably face the sun while others are in deep shadow, complicating radiometric balancing.
  • Standardization: There is a pressing need for better format standardization for 3D meshes and textured models to ensure interoperability between different NMA systems.
  • Automation: While LOD2 (roof structures) is becoming automated, reaching LOD3 (including windows and doors) automatically from aerial oblique data remains a significant research challenge.

Summary updated

BibTeX
@article{remondino2016oblique,
  author = {F. Remondino and I. Toschi and M. Gerke and F.C. Nex and D. Holland and A. McGill and J. Talaya Lopez and A. Magarinos},
  title = {{Oblique aerial imagery for NMA : some best practices}},
  journal = {ISPRS : Proceedings of the XXIII ISPRS Congress : From Human History to the Future With Spatial Information, 12-19 July , Prague, Czech Republic. Peer Reviewed Annals, Volume III-2, . Comm II, ThS14 Recent Developments in Open Data},
  year = {2016},
  url = {https://openreview.net/forum?id=zXdSReXP15},
}
Figure