Degenerate Motions in Multicamera Cluster SLAM with Non-overlapping Fields of View

TL;DR

This paper analyzes the conditions under which multicamera cluster SLAM systems with non-overlapping fields of view become degenerate, identifying specific motions that cause degeneracy and how increasing cameras reduces these problematic motions.

Contribution

It introduces a method to analyze degeneracy in multicamera SLAM by transforming the measurement Jacobian and identifies motions leading to degeneracy, also showing how more cameras mitigate this issue.

Findings

Degeneracy depends on relative motion and cluster configuration.

Increasing cameras reduces the set of degenerate motions.

Degeneracy can be analyzed via the rank of a reduced-dimension Jacobian matrix.

Abstract

An analysis of the relative motion and point feature model configurations leading to solution degeneracy is presented, for the case of a Simultaneous Localization and Mapping system using multicamera clusters with non-overlapping fields-of-view. The SLAM optimization system seeks to minimize image space reprojection error and is formulated for a cluster containing any number of component cameras, observing any number of point features over two keyframes. The measurement Jacobian is transformed to expose a reduced-dimension representation such that the degeneracy of the system can be determined by the rank of a dense submatrix. A set of relative motions sufficient for degeneracy are identified for certain cluster configurations, independent of target model geometry. Furthermore, it is shown that increasing the number of cameras within the cluster and observing features across different cameras over the two keyframes reduces the size of the degenerate motion sets significantly.