Registration of Laser Scanning Point Clouds: A Review

Liang Cheng, Song Chen, Xiaoqiang Liu, Hao Xu, Yang Wu, Manchun Li, Yanming ChenSensors 2018 · DOI

TL;DR

This paper provides a comprehensive review of LiDAR point cloud registration techniques, focusing on the prevalent coarse-to-fine strategy. It categorizes existing methods, highlights challenges like the lack of standardized datasets, and discusses future directions for integrating diverse LiDAR data sources for advanced geospatial applications.

What Problem Does This Paper Address?

The increasing demand for detailed 3D geospatial information has led to the widespread adoption of LiDAR technology. However, to create comprehensive 3D models and enable advanced applications (like environmental monitoring, urban planning, and disaster assessment), data from multiple LiDAR sources (different platforms, angles, and times) must be integrated. This integration hinges on accurate LiDAR data registration, the process of aligning these disparate point clouds. The challenge lies in the heterogeneity of LiDAR data (varying perspectives, resolutions, and noise levels) and the complexity of real-world environments, making robust and efficient registration a significant hurdle.

Methods

The paper primarily reviews existing methods, categorizing them within a coarse-to-fine registration strategy.

  • Coarse Registration: Aims to achieve a good initial alignment. This is typically done using feature-based methods, which rely on identifying and matching geometric primitives like points (e.g., building corners), lines (e.g., road networks, building contours), or surfaces (e.g., ground, roofs, facades). The paper discusses various point, line, and surface feature extraction and matching techniques.
  • Fine Registration: Refines the initial alignment to achieve maximum overlap and precision. Key methods reviewed include:
  • Iterative Approximation Methods: Primarily the Iterative Closest Point (ICP) algorithm and its numerous variants. ICP iteratively minimizes the distance between corresponding points in two point clouds.
  • Random Sample Consensus (RANSAC): A robust method for estimating model parameters from data containing outliers, widely used for point cloud registration due to its ability to handle noisy data and smaller overlaps.
  • Normal Distribution Transform (NDT): Converts point clouds into probability distribution functions within a grid, allowing for efficient and precise registration, especially for large datasets.
  • Methods with Auxiliary Data: Leverages additional information like images, GNSS coordinates, or standard targets to assist in registration.

The paper also details methods for error analysis, crucial for evaluating registration accuracy and comparing different techniques.

Datasets

The paper is a review and does not present new datasets. It references numerous studies that utilize various datasets, often from real-world scenarios like urban scenes, forest areas, and building reconstructions. The authors highlight the lack of standard datasets and unified evaluation systems as a significant limitation for objective comparison of registration methods.

Key Insights and Ideas

  • Coarse-to-Fine is King: The dominant paradigm for LiDAR registration, especially for complex scenes and multi-platform data, is a coarse-to-fine approach. This strategy effectively breaks down a difficult problem into manageable steps.
  • Feature Choice Matters: The success of coarse registration heavily relies on the quality and distribution of extracted features. While artificial objects offer more reliable geometric features, natural environments present challenges.
  • ICP's Ubiquity and Limitations: ICP is a foundational fine registration method, but its sensitivity to initial alignment and noise, along with computational cost for large datasets, necessitates improvements and complementary methods.
  • RANSAC and NDT Offer Robustness and Efficiency: RANSAC excels in handling outliers and partial overlaps, while NDT offers speed and precision for large-scale data, making them valuable for fine registration.
  • Heterogeneity is the Core Challenge: The inherent differences in data acquisition (perspective, resolution, noise) across various LiDAR systems are the primary drivers of registration complexity.
  • Need for Standardization: The absence of standardized benchmark datasets and evaluation metrics hinders objective comparison and progress in the field.

Key Results

As a review paper, it doesn't present novel experimental results. Instead, it synthesizes findings from numerous studies. The paper indicates that:

  • Feature-based methods are generally used for initial (coarse) registration, offering good results when features are well-defined and distributed.
  • ICP and its variants are widely used for fine registration, with improvements leading to significantly reduced mean offset distances (e.g., from 2.24 m to 0.12 m in one cited study).
  • RANSAC demonstrates good registration results even with small overlaps and is robust to noise.
  • NDT offers high computational speed and precision, particularly for large datasets, though it requires good initial alignment.
  • Auxiliary data methods (images, GNSS, targets) can simplify and improve registration, but their automation level varies.

The paper includes tables (e.g., Table 6 and 7) that summarize the applications, advantages, problems, and quantitative performance (deviations) of various registration methods based on the literature.

Limitations and Future Work

The authors identify several limitations and suggest future research directions:

  • Lack of Standardized Data and Evaluation: A critical limitation is the absence of unified benchmarks and evaluation systems, making objective comparison of methods difficult.
  • Challenges with Heterogeneous Data: Registration of multi-angle and multi-scale data from diverse LiDAR hardware (e.g., full-waveform LiDAR, new sensor types) remains an active area of research.
  • Indoor/Outdoor Integration: Extending registration techniques to seamlessly integrate indoor and outdoor environments is crucial for comprehensive 3D spatial understanding.
  • Large-Scale Registration Efficiency: As LiDAR datasets grow, improving the efficiency of registration algorithms, particularly for large-scale urban and global applications, is paramount. This includes optimizing data structures and indexing.
  • Objective Evaluation Systems: There is a strong need for objective evaluation systems, including standard datasets and comprehensive evaluation indices, to promote high-quality registration methods.
  • Application-Specific Needs: Future work should focus on developing registration methods tailored to specific applications like forest resource surveys, urban energy use, cultural heritage protection, and autonomous vehicles.

Summary updated

Abstract

The integration of multi-platform, multi-angle, and multi-temporal LiDAR data has become important for geospatial data applications. This paper presents a comprehensive review of LiDAR data registration in the fields of photogrammetry and remote sensing. At present, a coarse-to-fine registration strategy is commonly used for LiDAR point clouds registration. The coarse registration method is first used to achieve a good initial position, based on which registration is then refined utilizing the fine registration method. According to the coarse-to-fine framework, this paper reviews current registration methods and their methodologies, and identifies important differences between them. The lack of standard data and unified evaluation systems is identified as a factor limiting objective comparison of different methods. The paper also describes the most commonly-used point cloud registration error analysis methods. Finally, avenues for future work on LiDAR data registration in terms of applications, data, and technology are discussed. In particular, there is a need to address registration of multi-angle and multi-scale data from various newly available types of LiDAR hardware, which will play an important role in diverse applications such as forest resource surveys, urban energy use, cultural heritage protection, and unmanned vehicles.
BibTeX
@inproceedings{cheng2018registration,
  author = {Liang Cheng and Song Chen and Xiaoqiang Liu and Hao Xu and Yang Wu and Manchun Li and Yanming Chen},
  title = {{Registration of Laser Scanning Point Clouds: A Review}},
  booktitle = {Sensors},
  year = {2018},
  doi = {10.3390/s18051641},
}
Figure