[Paper Review] PCR-Pro: 3D Sparse and Different Scale Point Clouds Registration and Robust Estimation of Information Matrix For Pose Graph SLAM

M. Usman Maqbool Bhutta, Ming Liu|arXiv (Cornell University)|Aug 29, 2018

Robotics and Sensor-Based Localization4 citations

TL;DR

PCR-Pro proposes a robust method for 3D sparse and multi-scale point cloud registration that estimates accurate transformation and scale differences using keyframe-based pose refinement, achieving global convergence and minimal covariance for reliable information matrix estimation in pose graph SLAM.

ABSTRACT

For both indoor and outdoor environments, we propose an efficient and novel method for different scales and sparse 3D point clouds registration that cannot be handled by the current popular ICP approaches. Our algorithm efficiently detects the scale difference between point clouds and uses the keyframes to estimate the relative pose for calculating the scale difference. The algorithm applies a filter and computes the final transformation which coverages to a global minimum. The good estimation of transform and scale helps in the calculation of the covariance matrix using a closed form solution efficiently. This covariance between point clouds helps in the estimation of information matrix for pose-graph SLAM.

Motivation & Objective

• Address the challenge of registering sparse and multi-scale 3D point clouds that current ICP methods fail to handle effectively.
• Enable accurate transformation and scale estimation between point clouds without prior knowledge of scale differences.
• Achieve global convergence in point cloud registration to minimize estimation uncertainty.
• Provide a closed-form solution for covariance matrix estimation to support robust information matrix computation in pose graph SLAM.
• Support multi-agent SLAM systems by enabling fusion of data from diverse sensors and camera systems with inconsistent scale and sparsity.

Proposed method

• Uses keyframes to detect scale differences between source and target point clouds through relative pose estimation.
• Applies a filtering mechanism to refine correspondences and improve convergence to a global minimum.
• Employs an iterative closest point (ICP)-based optimization with a point-to-point error function to minimize transformation error.
• Estimates a 6×6 covariance matrix using the Hessian of the error function, derived from the second-order derivatives of the ICP objective.
• Computes the information matrix as the inverse of the estimated covariance matrix for use in pose graph optimization.
• Implements a threshold-based correspondence limit to reduce computation time while maintaining estimation accuracy.

Experimental results

Research questions

• RQ1 How can 3D point clouds with significant scale differences and sparsity be registered accurately when standard ICP fails?
• RQ2 Can a robust transformation and scale estimation method be developed that converges to a global minimum for sparse point clouds?
• RQ3 How can the covariance matrix of the transformation be efficiently and accurately estimated for use in pose graph SLAM?
• RQ4 What impact does low covariance have on the information matrix and subsequent pose graph optimization in SLAM?
• RQ5 Can the proposed method support multi-agent SLAM systems with heterogeneous sensor data and inconsistent scale?

Key findings

• PCR-Pro successfully converges to a global minimum in both indoor and outdoor environments, regardless of point cloud sparsity or scale differences.
• The estimated covariance matrix is consistently small due to high-accuracy transformation estimation, indicating low uncertainty in pose alignment.
• The information matrix derived from the inverse of the covariance matrix is highly reliable, allowing the use of an identity matrix with large weighting in pose graph optimization.
• The algorithm maintains low computational cost through a correspondence thresholding strategy, enhancing efficiency without sacrificing accuracy.
• Validation using LSD-SLAM datasets confirms robust performance across diverse real-world scenarios with varying scale and sparsity.
• The method enables effective fusion of point clouds from different sensors and systems, supporting multi-agent SLAM applications.