Assured Vision Geo-location with Precise Characterization of Errors

ABSTRACT: Qunav proposes the development of an assured vision geo-location (AVG) framework that supports accurate geo-registration and navigation while precisely characterizing their uncertainties. The framework adopts a multi-pose constrained estimation (MPCE) approach that ensures correct observability by decoupling landmark error states from motion states. Our AVG approach will support a Bayesian formulation with three methods, namely, outlier and clutter-suppression, outlier-estimation, and outlier-aware filtering. Each applies explicit modeling of non-Gaussian errors at a different level of processing. The Bayesian formulation generally involves nonlinear and non-Gaussian models that does not admit a closed form solution and as such will be implemented using a marginalized particle filter as an alternative to the extended Kalman filter. The main goal of Phase I is to demonstrate the technical feasibility of the proposed AVG solution. The Phase I development of observability-consistent and outlier-aware vision-based estimation methods will be focused on four main areas: 1) Modification of the MPCE method to address the observability issue and optimize performance; 2) Development and characterization of outlier/clutter detection and removal procedures for reliable estimation; 3) Development of methods for explicit modeling and estimation of measurement outliers; and, 4) Definition of performance metrics that faithfully characterize estimation errors and uncertainty reduction. BENEFIT: Phase I development will create a foundation for prototyping and transitioning of the technological approach for assured vision geo-location (AVG) with precise estimation of geo-location uncertainty. Successful accomplishment of Phase I tasks will enable to a) develop and verify all main algorithmic system components of the AVG framework; and, b) demonstrate the technical feasibility through extensive simulations and initial experimental validations. These anticipated results will ensure that a strong foundation is created for real-time technology demonstration during Phase II. The proposed AVG framework has a significant potential for both military and commercial applications. Example DOD applications include GPS-denied surveillance and reconnaissance missions where geo-location of objects of interests must be supplemented with estimated error bounds and estimation confidence levels. From the prospective of private sector, the largest commercialization potential will be achieved by transitioning the technology to automotive domain for cooperative vehicle safety applications.