Strip Adjustment of Airborne LiDAR Data in Urban Scenes Using Planar Features by the Minimum Hausdorff Distance

Ke Liu, Hongchao Ma, Liang Zhang, Zhan Cai, Haichi Ma2019 · DOI

TL;DR

This paper introduces an automated method for aligning overlapping airborne LiDAR strips in urban environments by matching building roof planes. By leveraging the Minimum Hausdorff Distance (MHD) for feature matching and a coarse-to-fine segmentation approach, the method effectively eliminates discrepancies between strips without manual intervention or external data, leading to more accurate 3D geospatial products.

What Problem Does This Paper Address?

Airborne LiDAR data is often collected in parallel strips. To create a seamless, comprehensive 3D map of an area, these strips must be stitched together. However, even with careful system calibration, slight misalignments (discrepancies) can occur between adjacent strips due to factors like imperfect calibration or operational procedures. These discrepancies can significantly impact the accuracy of derived products like Digital Elevation Models (DEMs). Traditional methods for aligning these strips, known as "strip adjustment," often rely on manual identification of corresponding features or require external data like ground control points or detailed maps (e.g., OpenStreetMap). This is labor-intensive, time-consuming, and not always feasible. This paper addresses the need for an automated, efficient, and accurate strip adjustment method specifically for urban scenes, where buildings offer abundant planar features.

Methods

The proposed method consists of four main steps:

  1. Building Extraction: Raw LiDAR point clouds are first filtered to separate ground from non-ground points. Then, semi-suppressed fuzzy C-means and restricted region growing algorithms are used to identify building points from the non-ground data.
  2. Building Segmentation and Matching: Building footprints are converted into a binary image. Connected components labeling is applied to segment individual buildings. Crucially, the Minimum Hausdorff Distance (MHD) is employed to match corresponding buildings between adjacent strips. MHD quantifies the similarity between two point sets by finding the maximum of the minimum distances between points in each set.
  3. Roof Plane Segmentation and Matching: For matched buildings, a coarse-to-fine strategy is used to segment roof planes. This involves calculating normal vectors and curvatures for points. Plane matching then uses a combination of MHD and normal vector similarity to identify corresponding roof planes between matched buildings. This step prioritizes larger planar patches for robustness and efficiency.
  4. Adjustment Computation: Once corresponding roof planes are identified, a rigid transformation model is used to estimate the rotation and translation parameters needed to align the strips. This is achieved by minimizing the Euclidean distance between the matched planes.

The key novelty lies in the automated detection and matching of planar features (building roofs) using MHD, avoiding manual selection and the need for external data.

Datasets

The paper evaluates the proposed method using two types of datasets:

  1. Single-channel LiDAR data: Acquired in the city of Niagara, this dataset consists of seven parallel strips. Six pairs of adjacent strips were used for evaluation.
  2. Dual-channel LiDAR data: Acquired by a Trimble AX80 system in Shijiazhuang, China, this dataset consists of two strips with near 100% overlap due to the dual-channel configuration.

The paper does not introduce new benchmark datasets but uses existing data to validate its method against established approaches.

Key Insights and Ideas

  • Leveraging Buildings for Alignment: Urban environments, with their abundance of buildings, provide a rich source of planar features (roofs) that are ideal for strip adjustment.
  • Minimum Hausdorff Distance for Robust Matching: MHD proves to be an effective metric for matching not only individual buildings but also their constituent roof planes, even in the presence of minor discrepancies.
  • Automated Feature Detection: The method significantly reduces manual effort by automatically detecting and matching corresponding planar features, a major bottleneck in traditional strip adjustment.
  • Coarse-to-Fine Segmentation: This strategy efficiently identifies relevant planar patches for adjustment by first segmenting broad areas and then refining the process.

Key Results

  • Effective Discrepancy Elimination: Experimental results demonstrate that the proposed method effectively eliminates discrepancies between adjacent LiDAR strips.
  • Improved Accuracy: The method achieved slightly better adjustment results compared to a comparison method (Wu and Fan [17]) and commercial software (TerraMatch), as indicated by lower average distances (Dmean) and higher normalized RMSE (NRMSE) improvements. For instance, in the single-channel LiDAR experiment, the NRMSE improvement was 0.8% over the comparison method.
  • Automatic Operation: The method operates in a more automatic manner than comparison methods, requiring no manual operations or external data.
  • Dual-Channel LiDAR Performance: The method also showed effectiveness for dual-channel LiDAR data, achieving a Dmean reduction from 0.258 m to 0.009 m.

Limitations and Future Work

  • Parameter Sensitivity: The coarse-to-fine roof segmentation requires several predefined parameters (e.g., k, C50, D50, Dpo). While optimal values were determined through experimentation, this limits the full automation level.
  • Preference for Planar Features: The method relies heavily on planar features. While effective for buildings, it might be less suitable for areas with limited or complex building structures.
  • Selection of Corresponding Planes: The paper notes that while many corresponding planes are detected, selecting planes with similar orientations and slopes might be unnecessary and could potentially reduce accuracy. The authors suggest that planes with different orientations and slopes are preferred for better correlation among estimated parameters.
  • Future Work: The authors suggest further research into developing more automatic methods for planar patch detection, exploring other distance metrics beyond Euclidean distance for Hausdorff matching, and investigating the impact of selecting planes with diverse orientations and slopes for improved adjustment accuracy. They also highlight the potential for further refinement by minimizing discrepancies observed when using different strip-pair orders for adjustment.

Summary updated

BibTeX
@misc{liu2019strip,
  author = {Ke Liu and Hongchao Ma and Liang Zhang and Zhan Cai and Haichi Ma},
  title = {{Strip Adjustment of Airborne LiDAR Data in Urban Scenes Using Planar Features by the Minimum Hausdorff Distance}},
  year = {2019},
  doi = {10.3390/s19235131},
}

Loading related content...

Figure