Vision-Based State Estimation for Autonomous Micro Air Vehicles

The realization of vision-based, autonomousflight formicro air vehicles presents numerous technical challenges in diverse fields such as image processing, trajectory planning, control theory, and microhardware design. This paper explores one aspect of the overall problem: robust, real-time estimation of aircraft states froma set of tracked feature points. This vision-based approach is realized via an implementation of the implicit extended Kalman filter, a variation of the classical Kalman filter that allows for nonlinear dynamic models and updates from measurements that are implicit functions of the state variables. The epipolar constraint, which is a geometric relationship between the feature point position vectors and the camera translation vector, is employed in this paper as the implicit measurement in the Kalman filter. In order for the vison-based Kalman filter to provide reliable state estimation, it must be robust with respect to modeling errors in the dynamic propagator. This is especially true because the aerodynamic forces andmoments acting onamicro air vehicle are frequently not known to ahighdegree of accuracy. Furthermore, the algorithmmust be robust with respect to noise in themeasured feature point positions in the image plane. This paper explores both of these robustness issues using results from a micro air vehicle simulation model developed at the NASA Langley Research Center. In particular, a hierarchy of dynamic models, ranging from a random walk model to a high-fidelity nonlinear micro air vehicle model, is employed in the Kalman filter for a simulated micro air vehicle trajectory with varying levels of measurement noise. It is demonstrated that the vision