CorAl -- Are the point clouds Correctly Aligned?

TL;DR

CorAl is a novel method for automatically assessing the alignment quality of point cloud pairs in robotics perception, effectively detecting small misalignments across diverse environments without environment-specific parameters.

Contribution

It introduces CorAl, an entropy-based measure and classifier for point cloud registration quality, improving detection accuracy of small misalignments in unseen environments.

Findings

CorAl detects small alignment errors with 95% accuracy.

It outperforms previous methods in diverse environments.

CorAl is sensitive to minor misalignments due to entropy comparison.

Abstract

In robotics perception, numerous tasks rely on point cloud registration. However, currently there is no method that can automatically detect misaligned point clouds reliably and without environment-specific parameters. We propose "CorAl", an alignment quality measure and alignment classifier for point cloud pairs, which facilitates the ability to introspectively assess the performance of registration. CorAl compares the joint and the separate entropy of the two point clouds. The separate entropy provides a measure of the entropy that can be expected to be inherent to the environment. The joint entropy should therefore not be substantially higher if the point clouds are properly aligned. Computing the expected entropy makes the method sensitive also to small alignment errors, which are particularly hard to detect, and applicable in a range of different environments. We found that CorAl is able to detect small alignment errors in previously unseen environments with an accuracy of 95% and achieve a substantial improvement to previous methods.