QuNav’s vision-aided inertial solution maintains robust and consistent navigation performance in GPS-denied scenarios.

QuNav’s vision-aided inertial mechanization supports:

• Monocular video images

• Higher-grade and consumer-grade MEMS inertial

• Ground vehicle and UAV applications;

Imaging systems are increasingly used in various applications where they serve not only the purpose of object detection and recognition but also as a navigation aid. A key challenge of vision-aided navigation is consistency of its estimation performance. First, the assumption of a Gaussian distribution for measurement errors may not be valid due to outliers commonly resulting from complicated algorithmic processing of images (e.g., feature extraction, feature matching, and frame-to-frame tracking). Second, the conventional extended Kalman filter (EKF) is known to be optimistic in its estimation error covariance. The inconsistency stems from under-estimation of nonlinear errors and may be attributed to non-observability of some system states.

To enable consistent performance of vision-aided navigation, QuNav has developed a real-time prototype of Assured Vision-Aided Inertial Localization (AVAIL). AVAIL mitigates outliers in vision measurements with probabilistic data association filtering that:

• Computes a probability of an outlier missed-detection by statistical gating, and

• De-weights vision measurements accordingly.

Correct observability properties are enabled by a batch estimator that:

• Processes feature tracks (rather than individual vision measurements),

• Separates INS and feature error states (via a null-state projection), and

• Applies inverse-depth parameterization of monocular vision measurements (to mitigate linearization errors).

Consistent estimation performance of AVAIL was successfully demonstrated for:

• Ground vehicle and UAV test scenarios;

• Higher-grade MEMS and consumer-grade MEMS inertial sensors aided by monocular video images.