Answers are grounded in this paper and cite their sources.
Integrity monitoring (IM) systems assess the trustworthiness of navigation output in real time and are indispensable to the advancement of connected and autonomous vehicles (CAVs) toward full autonomy. Although existing IM methods attempt to assess heterogeneous vision-based navigation systems in real time, they lack a unified account of the error propagation pathways inherent in the visual front-end. Moreover, visual systems are affected by multiple error sources, including those arising from the imaging pipeline, the operating environment, and the front-end processing modules. These error conditions frequently violate the commonly assumed identically distributed Gaussian pixel error model adopted in many IM frameworks. To address these two issues, this paper first introduces a unifying construct termed the error-space. It characterizes how errors propagate and couple within test statistics, provides a common framework for the consistency-checking mechanisms used in both IM and visual IM methods, and clarifies the intrinsic link between error sources and fault separability. Building on this analysis, the paper further proposes two IM algorithms for visual systems that are both based on feature pairs: Optical Flow-based Point-Pair Integrity Monitoring (OPIM) and Martingale-Majority-Certificate Optical Flow-based Point-Pair Integrity Monitoring (MMC–OPIM). Among them, MMC–OPIM leverages a majority certificate mechanism to eliminate reliance on specific error distribution assumptions. It offers a more comprehensive solution for vision IM under real-world conditions. Experimental results confirm that the proposed OPIM and MMC–OPIM algorithms outperform existing visual IM methods in integrity monitoring on image sequences.
No takes yet. Share an insight, caveat, or question.
Zhao et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: