Vision-Aided Inertial Navigation with Modeling of Measurement Outliers

For vision-based navigation applications, the assumption of a Gaussian distribution for measurement errors may not be valid due to outliers commonly resulting from complicated algorithmic processing of images (for example, feature extraction, feature matching, and frame-to-frame tracking). Although many algorithms have been developed to minimize the probability of outputs that are outliers, the probability is still nonzero and is not estimated with current approaches. This paper develops a vision-aided inertial navigation mechanization that explicitly accounts for the presence of outliers in vision measurements. Navigation mechanization is augmented by probabilistic data association filtering. Probabilistic data association filtering takes into account nonperfect detection of outliers by adaptively computing the probability that an outlier is undetected and weighting vision measurements accordingly. Simulation and experimental results are used to demonstrate that the probabilistic data association filteri...