Consistency analysis and improvement for single-camera localization

In this paper, we study the problem of estimator inconsistency in single-camera simultaneous localization and mapping (MonoSLAM) from a standpoint of system observability. Specifically, we postulate that a leading cause of inconsistency is the gain of spurious information along unobservable directions, resulting in smaller uncertainties, larger estimation errors, and divergence. Moreover, we introduce an Observability-Constrained MonoSLAM (OC-MonoSLAM) approach, which explicitly enforces the unobservable directions of the system, hence preventing spurious information gain and reducing inconsistency. Our analysis, along with the proposed method for reducing inconsistency, are validated with simulation trials and real-world experimentation.