Generalized Optimal-State-Constraint Extended Kalman Filter (OSC-EKF)

Cameras, inertial measurement units IMUs, and computational power have all become practical, and the demand for small and efficient navigation systems that do not rely on external infrastructure such as GPS is high. Combining visual information with inertial sensing is a challenging problem. The optimal-state-constraint extended Kalman filter OSC-EKF is a new method previously designed to optimally combine relative pose constraints from a monocular camera with the output of an IMU. This framework is generalized so that any combination of sensors that can be combined to produce relative pose constraints can be used to update the EKF. A stereo vision-structure and motion SAM problem and a monocular SAM problem are both used to update the OSC-EKF without making any changes to the EKF. The efficacy of these algorithms is demonstrated by achieving reasonable consistency and accuracy on a challenging micro aerial vehicle dataset.