High-Precision Registration of Point Clouds Based on Sphere Feature Constraints

Junhui Huang, Wang Zhao, Jianmin Gao, Youping Huang, David P. TowersSensors 2016 · DOI

TL;DR

This paper introduces a novel "Sphere Feature Constraint Registration" (FCR) method to achieve highly precise 3D point cloud registration, even for challenging datasets with non-overlapping areas or uniform surfaces. By strategically using sphere targets to create "virtual overlapping areas" and incorporating a weighted optimization approach, FCR significantly improves accuracy and robustness compared to existing methods.

What Problem Does This Paper Address?

Accurate 3D measurement relies heavily on point cloud registration, where multiple scans of an object are aligned into a single, complete point cloud. However, traditional registration methods struggle with datasets that lack sufficient overlapping areas or have surfaces with little curvature variation (like planes or spheres). These limitations lead to imprecise alignments, hindering the overall accuracy of 3D measurements. Existing methods either require significant overlap (free registration like ICP) or rely on external references that can introduce their own precision limitations (reference-based registration).

Methods

The FCR method tackles these challenges through a two-pronged approach:

  1. Constructing Virtual Overlapping Areas: The core innovation is the use of pre-placed sphere targets. These targets, with known radii, are used to fit ideal sphere surfaces. By aligning the measurement point clouds to these constructed surfaces, the method effectively creates "virtual overlapping areas." This bypasses the need for actual geometric overlap in the raw data and provides a clean, continuous reference surface. The sphere centers are identified using Hough transforms and then roughly registered using Sphere Center Registration (SCR).
  2. Optimized Transformation Solution: Once corresponding points are identified between the measurement point clouds and the virtual overlapping areas, a refined registration is performed. The paper employs an optimization method (similar to ICP) to solve for the rotation (R) and translation (T) matrices. Crucially, a weight function is incorporated into the cost function. This weighting scheme helps to reduce the impact of large noise in the point clouds, thereby improving the stability and accuracy of the final registration.

Datasets

The paper uses simulated datasets for evaluation, generated with varying parameters such as target size, point density, noise levels, and overlap ratios. Real experimental data from a standard ball and a turbine blade are also used to validate the method's performance in practical scenarios. No new benchmark datasets were introduced.

Key Insights and Ideas

  • Virtual Overlap is Key: The concept of creating "virtual overlapping areas" using known geometric primitives (spheres) is a powerful way to overcome the limitations of real-world data scarcity in overlapping regions.
  • Sphere Targets as Constraints: Sphere targets are not just for rough alignment; their constrained, ideal surfaces provide a robust reference for fine registration, reducing noise and preventing convergence to local optima.
  • Weighted Optimization for Robustness: Incorporating a weight function in the optimization process significantly enhances the method's resilience to noise, a common issue in 3D scanning.
  • Universality of Features: The paper suggests that other geometric features (like cones) could potentially be used, hinting at the broader applicability of the FCR principle.

Key Results

  • High Precision: The FCR method achieved a mean registration error of 3.3 µm with a standard deviation of 1.7 µm in simulations, outperforming SCR (11.4 µm mean error) and ICP (22.4 µm mean error with overlapping data, failure with non-overlapping data).
  • Robustness to Non-Overlapping Data: FCR demonstrated consistent performance whether overlapping data was present or not, a significant advantage over ICP.
  • Insensitivity to Parameters: The method showed minimal sensitivity to target size, point density, and overlap ratio, with performance degradation only becoming significant with high mean error (systematic bias) in the input data.
  • Real-World Validation: Experimental results on a standard ball and a turbine blade confirmed FCR's superior precision compared to SCR and ICP, with mean errors of 0.012 mm and 0.020 mm respectively for the ball and plate.

Limitations and Future Work

  • Mean Error Impact: The paper highlights that a significant mean error (systematic bias) in the input point clouds can negatively impact registration precision, especially without overlapping data. The authors note that this systematic error can often be compensated for beforehand.
  • Computational Cost: While FCR's speed is comparable to ICP, it is noted as being slower than SCR.
  • Future Work: The paper suggests that other feature targets (e.g., cones) could be used, implying potential for extending the FCR framework to different geometric primitives. Further exploration of optimizing the weight function and its impact could also be a direction.

Summary updated

Abstract

Point cloud registration is a key process in multi-view 3D measurements. Its precision affects the measurement precision directly. However, in the case of the point clouds with non-overlapping areas or curvature invariant surface, it is difficult to achieve a high precision. A high precision registration method based on sphere feature constraint is presented to overcome the difficulty in the paper. Some known sphere features with constraints are used to construct virtual overlapping areas. The virtual overlapping areas provide more accurate corresponding point pairs and reduce the influence of noise. Then the transformation parameters between the registered point clouds are solved by an optimization method with weight function. In that case, the impact of large noise in point clouds can be reduced and a high precision registration is achieved. Simulation and experiments validate the proposed method.
BibTeX
@inproceedings{huang2016high,
  author = {Junhui Huang and Wang Zhao and Jianmin Gao and Youping Huang and David P. Towers},
  title = {{High-Precision Registration of Point Clouds Based on Sphere Feature Constraints}},
  booktitle = {Sensors},
  year = {2016},
  doi = {10.3390/s17010072},
}
Figure