Feature matching error analysis and modeling for consistent estimation in vision‐aided navigation

Abstract

In vision-aided navigation, images from natural scenes are processed to produce feature measurements for navigation aiding. However, feature measurements are subject to non-Gaussian errors and, in particular, to outliers in feature extraction and matching. If not properly accounted for, the errors are likely to lead to inconsistent, usually optimistic, estimation. To model feature matching errors, we study the circumstances in which outliers occur by checking the extracted features and detected outliers against the original images so as to verify spatial and temporal assumptions about the feature errors and their distributions. These error models are used in the probabilistic data association filter (PDAF) that updates an inertial navigation solution with feature measurements using the probability that an outlier is undetected and the corresponding feature measurement and the estimation error covariance are weighted down accordingly. Ground vehicle data show consistent estimation of this approach in the presence of real-world image processing errors.