|M.Sc Student||Maxim Goldshtein|
|Subject||Computer Vision-Based Compensation of Gyro Bias and Scale-|
|Department||Department of Aerospace Engineering||Supervisors||Full Professor Oshman Yaakov|
|Dr. Efrati Tzvi|
This thesis presents an online algorithm for computer vision-based estimation of bias and scale-factor of rate gyros used for inertial stabilization of a tracking camera. The design of modern autonomous systems is required to be cost-effective, calling for the use of micro-electromechanical sensors (MEMS). Because these sensors are prone to large bias and scale-factor errors, these errors must be compensated for by using an estimator that utilizes, in the present case, vision data. This research focuses on an image-guided missile, where the visual information is acquired by an inertially-stabilized tracking camera installed on a gimbaled system with two stabilizing MEMS gyros. An additional roll-damping gyro is installed on the missile body.
The algorithm uses the optical flow, extracted from a sequence of tracker-acquired images, the gyro measurements, gimbal angle sensors, missile control fins and a partial missile aerodynamic model. The visual information is combined with the measurements by way of an implicit constraint, which is used in an Implicit Constraint Extended Kalman Filter (ICEKF) for bias and scale-factor estimation. A new derivation of the ICEKF is presented, that highlights possible numerical problems and proposes methods for their alleviation.
Three approaches for creating the implicit-constraint functions are investigated: an extension of the existing subspace-constraint method, a simplified version of the existing method, and a new, geometry-based approach. The new method is simpler, more computationally-efficient and less hardware-demanding than previous approaches, without affecting the estimation performance. The method is independent of the scene's structure and facilitates outlier rejection.
The algorithm's performance is assessed using extensive simulations, comparing the performance of the methods and checking the robustness of the new method. It is shown that the algorithm enables achieving good miss-distance results, if the missile is using MEMS gyros in camera stabilization, guidance and autopilot.