Multifaceted Geometric Assessment towards Simplified Urban Surfaces Built by 3D Reconstruction

Sheng'en Liu, Hui Yi, Xiangning Chen, Decheng Wang, Wei JinISPRS Int. J. Geo Inf 2019 · DOI

TL;DR

This paper introduces a novel, multifaceted approach to assess the quality of simplified 3D urban surfaces generated from aerial imagery. By classifying urban surfaces into planar, detailed, and framework categories, the authors provide a more nuanced evaluation than existing methods, enabling better understanding and improvement of 3D reconstruction and simplification algorithms.

What Problem Does This Paper Address?

Generating detailed 3D models of urban environments from aerial photographs is crucial for applications like smart cities, urban planning, and disaster relief. However, these models often contain massive amounts of data, necessitating simplification for efficient processing, storage, and real-time visualization. The challenge lies in assessing the quality of this simplification. Traditional methods, like the Metro tool, often fail to capture the nuances of urban surfaces, particularly the preservation of important geometric features like building edges (urban frameworks) and intricate details. This lack of detailed assessment makes it difficult to compare different simplification algorithms effectively and to identify areas for improvement. The paper addresses this by proposing a more comprehensive assessment framework that accounts for the heterogeneous nature of urban environments.

Methods

The core of this paper is a novel assessment methodology that begins with surface classification. They first identify planar vertices by analyzing the convergence of points to a plane using a fluctuation parameter ($\epsilon_f$) derived from eigenvalues of covariance matrices. This is followed by extracting planar surfaces using a refined method that combines dihedral angle analysis with planar vertex recognition and an extension step.

Once surfaces are classified into planar, detailed, and urban frameworks, the paper introduces a multifaceted sampling strategy. For detailed surfaces, sampling points are weighted to emphasize borders and virtual faces (representing architectural edges). For urban frameworks, sampling is performed on edges, with weights adjusted to highlight curvature and corners.

Finally, a set of assessment indexes are proposed:

  • Mean error of planar surfaces: Measures deviation from the sampled surface.
  • Planarity indicator ($I_p$): Quantifies how well planar surfaces are preserved.
  • Proportion of faces of detailed surfaces ($SQ\%$): Indicates the density of geometric elements on detailed areas.
  • Error disparity of detailed surfaces ($\Delta E_q$): Highlights areas where simplification errors are significant relative to comprehensive error, suggesting room for improvement.
  • Gap between errors from different simplification algorithms ($\Delta E_{\alpha\beta}$): Allows for fine-grained comparison of algorithms.

Datasets

The paper utilizes two datasets, referred to as Scenery 1 and Scenery 2, which are described as input surfaces from aerial photographs. These datasets are used to test and validate the proposed assessment methods and algorithms. The paper also refers to "tile A" and "tile B" in the experimental results, which appear to be subsets or specific regions of the larger scenery datasets.

Key Insights and Ideas

  • Urban surfaces are heterogeneous: A one-size-fits-all simplification assessment is insufficient. Differentiating between planar, detailed, and framework elements is crucial.
  • Urban frameworks (edges/borders) are critical: These features are vital for semantic recognition and applications like urban planning, and their preservation is paramount.
  • Existing tools (like Metro) are limited: They often fail to capture errors in critical geometric features and provide a superficial assessment.
  • A multifaceted approach is necessary: Combining surface classification with targeted sampling and a suite of specialized indexes provides a deeper understanding of simplification quality.
  • Error disparity ($\Delta E_q$) is a powerful indicator: It goes beyond simple error measurement to pinpoint areas where simplification algorithms can be improved.

Key Results

  • The proposed method successfully classifies urban surfaces into planar, detailed, and urban frameworks.
  • Experiments show that the proposed assessment indexes provide a more nuanced understanding of simplification quality compared to Metro.
  • The paper demonstrates that certain algorithms, like aQSlim, perform better in preserving detailed surfaces, while others, like DVDC ($\lambda=2$), excel in maintaining the planarity of planar surfaces.
  • The study suggests that combining different simplification strategies (e.g., remeshing followed by edge collapse) can lead to improved overall results.
  • The mean error of urban frameworks proved to be a more stable indicator for determining optimal parameters for simplification algorithms compared to detailed surfaces.
  • The proposed method's time performance was found to be comparable to or better than Metro.

Limitations and Future Work

  • Sensitivity to outliers: The initial method for extracting planar vertices based on dihedral angles was found to be sensitive to outliers, necessitating the development of a more robust approach.
  • Experimental parameter tuning: While the paper explores optimal parameters for some algorithms, the process of determining these parameters can be complex and may require further refinement.
  • Future work: The authors suggest exploring objective assessments of newly developed simplification algorithms and investigating a wider range of algorithm combinations to develop improved simplification methods for urban models. They also highlight the need for further research into the optimal choice of curvature indicators for specific urban framework simplification scenarios.

Summary updated

BibTeX
@article{liu2019multifaceted,
  author = {Sheng'en Liu and Hui Yi and Xiangning Chen and Decheng Wang and Wei Jin},
  title = {{Multifaceted Geometric Assessment towards Simplified Urban Surfaces Built by 3D Reconstruction}},
  journal = {ISPRS Int. J. Geo Inf.},
  year = {2019},
  doi = {10.3390/ijgi8080360},
}
Figure