Measurement Error Detection for Stereo Visual Odometry Integrity

Abstract Integrity, a safety-of-life framework from civil aviation for satellite navigation, is greatly under-explored in visual navigation. A new two-factor approach to rejecting measurement outliers proposed navigation integrity stereo odometry (VO). In contrast other treatments using reprojection error as residuals, our choice of landmark matching inherently connects solutions and monitoring. We propose two methods detect large residuals that cannot otherwise be identified by existing outlier rejection state-of-the-art VO pipelines. By these outliers, can bounded the distribution overbounding method provides fundamental inputs computations. evaluate an open-source data set. Overbounding performance improved terms tightness, computational efficiency, most important all, scenario tolerance. This could good starting point developing future monitoring algorithms particular, VO.