Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DeepAAT: Deep Automated Aerial Triangulation for Fast UAV-based Mapping

Zequan Chen, Jianping Li, Qusheng Li, Bisheng Yang, Zhen Dong

TL;DR

DeepAAT addresses the efficiency and robustness gaps in UAV-based Automated Aerial Triangulation by proposing a dedicated deep network that fuses spatial, spectral, and GPS information. The Spatial-Spectral Feature Aggregation module, Global Consistency-based Outlier Rejecting module, and Pose Decode module jointly yield accurate pose predictions and reliable inlier scoring, enabling large-scale, GPU-bounded processing via a divide-and-conquer workflow with cluster merging. Empirical results demonstrate that DeepAAT substantially accelerates AAT (hundreds of times faster than incremental methods and tens of times faster than global methods) while achieving comparable reconstruction accuracy, with strong robustness to outliers and good generalization across varying image set sizes. The approach promises practical impact for scalable, high-quality UAV 3D mapping and photogrammetry workflows, with code released to the community.

Abstract

Automated Aerial Triangulation (AAT), aiming to restore image pose and reconstruct sparse points simultaneously, plays a pivotal role in earth observation. With its rich research heritage spanning several decades in photogrammetry, AAT has evolved into a fundamental process widely applied in large-scale Unmanned Aerial Vehicle (UAV) based mapping. Despite its advancements, classic AAT methods still face challenges like low efficiency and limited robustness. This paper introduces DeepAAT, a deep learning network designed specifically for AAT of UAV imagery. DeepAAT considers both spatial and spectral characteristics of imagery, enhancing its capability to resolve erroneous matching pairs and accurately predict image poses. DeepAAT marks a significant leap in AAT's efficiency, ensuring thorough scene coverage and precision. Its processing speed outpaces incremental AAT methods by hundreds of times and global AAT methods by tens of times while maintaining a comparable level of reconstruction accuracy. Additionally, DeepAAT's scene clustering and merging strategy facilitate rapid localization and pose determination for large-scale UAV images, even under constrained computing resources. The experimental results demonstrate DeepAAT's substantial improvements over conventional AAT methods, highlighting its potential in the efficiency and accuracy of UAV-based 3D reconstruction tasks. To benefit the photogrammetry society, the code of DeepAAT will be released at: https://github.com/WHU-USI3DV/DeepAAT.

DeepAAT: Deep Automated Aerial Triangulation for Fast UAV-based Mapping

TL;DR

DeepAAT addresses the efficiency and robustness gaps in UAV-based Automated Aerial Triangulation by proposing a dedicated deep network that fuses spatial, spectral, and GPS information. The Spatial-Spectral Feature Aggregation module, Global Consistency-based Outlier Rejecting module, and Pose Decode module jointly yield accurate pose predictions and reliable inlier scoring, enabling large-scale, GPU-bounded processing via a divide-and-conquer workflow with cluster merging. Empirical results demonstrate that DeepAAT substantially accelerates AAT (hundreds of times faster than incremental methods and tens of times faster than global methods) while achieving comparable reconstruction accuracy, with strong robustness to outliers and good generalization across varying image set sizes. The approach promises practical impact for scalable, high-quality UAV 3D mapping and photogrammetry workflows, with code released to the community.

Abstract

Automated Aerial Triangulation (AAT), aiming to restore image pose and reconstruct sparse points simultaneously, plays a pivotal role in earth observation. With its rich research heritage spanning several decades in photogrammetry, AAT has evolved into a fundamental process widely applied in large-scale Unmanned Aerial Vehicle (UAV) based mapping. Despite its advancements, classic AAT methods still face challenges like low efficiency and limited robustness. This paper introduces DeepAAT, a deep learning network designed specifically for AAT of UAV imagery. DeepAAT considers both spatial and spectral characteristics of imagery, enhancing its capability to resolve erroneous matching pairs and accurately predict image poses. DeepAAT marks a significant leap in AAT's efficiency, ensuring thorough scene coverage and precision. Its processing speed outpaces incremental AAT methods by hundreds of times and global AAT methods by tens of times while maintaining a comparable level of reconstruction accuracy. Additionally, DeepAAT's scene clustering and merging strategy facilitate rapid localization and pose determination for large-scale UAV images, even under constrained computing resources. The experimental results demonstrate DeepAAT's substantial improvements over conventional AAT methods, highlighting its potential in the efficiency and accuracy of UAV-based 3D reconstruction tasks. To benefit the photogrammetry society, the code of DeepAAT will be released at: https://github.com/WHU-USI3DV/DeepAAT.
Paper Structure (28 sections, 10 equations, 11 figures, 8 tables)

This paper contains 28 sections, 10 equations, 11 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Classic Automated Aerial Triangulation
  3. Supervised Automated Aerial Triangulation
  4. Preliminary
  5. Problem Definition of Automated Aerial Triangulation
  6. Projective Factorization
  7. Permutation Equivariant Layer
  8. System Overview
  9. Network Architecture of DeepAAT
  10. Spatial-Spectral Feature Aggregation Module
  11. Global Consistency-based Outlier Rejecting Module
  12. Pose Decode Module
  13. Loss Function
  14. Experiments
  15. Dataset
  16. ...and 13 more sections

Figures (11)

  • Figure 1: Exchangeability of tensor $\mathbf{W}$.
  • Figure 2: System overview of the efficient UAV-based mapping system.
  • Figure 3: Network architecture of DeepAAT.
  • Figure 4: UAV-based image dataset used for the experiments. The dataset is divided into eight blocks.
  • Figure 5: Image clustering result, where the number after "_" represents the number of cameras included in the subset.
  • ...and 6 more figures