Assured Vision Geo-location with Precise Characterization of Errors

ABSTRACT: Phase I effort has developed the framework of Assured Vision-Aided Inertial Localization (AVAIL) and demonstrated its feasibility with simulated and experimental data. Phase II will extend AVAIL into a Geo-registration and Image-Aided Navigation Technology (GIANT). The ultimate goal of Phase II is to implement a real-time breadboard prototype of GIANT and demonstrate its operation on a ground vehicle and a small unmanned aerial system. GIANT will maintain consistent observability properties by utilizing a batch-based multi-pose constrained estimation (MPCE) that 1) separates ownship and landmark navigation states via a null-space projection; and, 2) uses the inverse depth parameterization of landmarks to significantly increase the linearity region. Outliers in vision measurement will be mitigated with probabilistic data association filtering (PDAF). PDAF takes into account non-perfect detection of outliers by adaptively computing a probability that an outlier is undetected by statistical gating and weighting vision measurements accordingly. For the extension to geo-registration, MPCE will be augmented with an unscented transform that estimates landmark states using a robust cost function for mitigation of outliers. Phase II will fully develop the GIANT technological approach, will evaluate its performance with experimental data, and will demonstrate system functionality in representative GPS-denied ground and flight environments in real-time. BENEFIT: When successfully demonstrated, GIANT has a significant potential for both DoD and civilian markets. For military applications, GIANT technology is directly applicable to precision guided munitions and autonomous aerial vehicles by enabling accurate and robust navigation and guidance capabilities for GPS-denied scenarios. In addition, accurate characterization of navigation and geo-location uncertainty will enable deployment of computer-vision based algorithms for civilian applications that require known uncertainties. One particular application is the vehicular collaborative navigation for the intelligent transportation systems. For this application, GIANT will produce the vehicle position estimates and in particular their uncertainties that are critical parameters to generate driver-alert messages and directives in order to avoid property damages and injuries from collisions.