Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Sight View Constraint for Robust Point Cloud Registration

Yaojie Zhang, Weijun Wang, Tianlun Huang, Zhiyong Wang, Wei Feng

TL;DR

This paper proposes a novel and general Sight View Constraint (SVC) to conclusively identify incorrect transformations, thereby enhancing the robustness of existing PCR methods and highlighting the significance of the decision version problem of partial PCR.

Abstract

Partial to Partial Point Cloud Registration (partial PCR) remains a challenging task, particularly when dealing with a low overlap rate. In comparison to the full-to-full registration task, we find that the objective of partial PCR is still not well-defined, indicating no metric can reliably identify the true transformation. We identify this as the most fundamental challenge in partial PCR tasks. In this paper, instead of directly seeking the optimal transformation, we propose a novel and general Sight View Constraint (SVC) to conclusively identify incorrect transformations, thereby enhancing the robustness of existing PCR methods. Extensive experiments validate the effectiveness of SVC on both indoor and outdoor scenes. On the challenging 3DLoMatch dataset, our approach increases the registration recall from 78\% to 82\%, achieving the state-of-the-art result. This research also highlights the significance of the decision version problem of partial PCR, which has the potential to provide novel insights into the partial PCR problem.

Sight View Constraint for Robust Point Cloud Registration

TL;DR

This paper proposes a novel and general Sight View Constraint (SVC) to conclusively identify incorrect transformations, thereby enhancing the robustness of existing PCR methods and highlighting the significance of the decision version problem of partial PCR.

Abstract

Partial to Partial Point Cloud Registration (partial PCR) remains a challenging task, particularly when dealing with a low overlap rate. In comparison to the full-to-full registration task, we find that the objective of partial PCR is still not well-defined, indicating no metric can reliably identify the true transformation. We identify this as the most fundamental challenge in partial PCR tasks. In this paper, instead of directly seeking the optimal transformation, we propose a novel and general Sight View Constraint (SVC) to conclusively identify incorrect transformations, thereby enhancing the robustness of existing PCR methods. Extensive experiments validate the effectiveness of SVC on both indoor and outdoor scenes. On the challenging 3DLoMatch dataset, our approach increases the registration recall from 78\% to 82\%, achieving the state-of-the-art result. This research also highlights the significance of the decision version problem of partial PCR, which has the potential to provide novel insights into the partial PCR problem.
Paper Structure (25 sections, 1 theorem, 14 equations, 4 figures, 5 tables, 2 algorithms)

This paper contains 25 sections, 1 theorem, 14 equations, 4 figures, 5 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Model Fitting
  4. RANSAC-based methods
  5. Spatial compatibility methods
  6. Metrics for Model fitting
  7. Method
  8. Reminder of the PCR task
  9. Sight view Constraint
  10. The SVC-based algorithm
  11. Experiments and Results
  12. Datasets and Experimental Setup
  13. Datasets
  14. Evaluation Criteria
  15. Implementation Details
  16. ...and 10 more sections

Key Result

Theorem 1

In a static environment, the transformed source point cloud cannot block the line of sight between the target point cloud and the sensor. Otherwise, the estimated transformation is incorrect.

Figures (4)

  • Figure 1: An illustration of the core idea of SVC. The line (green or red) represents the sight view line from the sensor viewpoint to the target point. The red line means there exist source points in the line so that the sight view is blocked. The green line means the sight view is not blocked. If there exist lots of red lines, then the estimated transformation is wrong.
  • Figure 2: A simple pipeline of the correspondence-based PCR method. (1) Generate initial correspondences according to feature descriptors. (2) Identify inliers and outliers to generate multiple transformation hypotheses. (3) Using SVC-based evaluation to select the optimal transformation. (4) Merge point clouds according to the transformation.
  • Figure 3: Time efficiency analysis. Fig.\ref{['fig:time-a']} is about one single execution of the SVC algorithm. Fig.\ref{['fig:time-b']} is about the average time of the SC2 combined with SVC on different datasets.
  • Figure 4: Registration recall changes with different numbers of estimated hypotheses.

Theorems & Definitions (2)

  • Theorem 1
  • proof