A fusion strategy for reliable attitude measurement using MEMS gyroscope and camera during discontinuous vision observations

For indoor close-range high-accuracy attitude measurement using inertial and vision sensors, the major challenge is to achieve a reliable attitude in various scenarios, specifically in the absence of vision information. To address this challenge, we present a fusion strategy for reliable attitude measurement using a micro electro mechanical system (MEMS) gyroscope and camera, particularly during discontinuous vision observations. The proposed algorithm consecutively executes update and prediction modes. In the update mode, when both inertial and visual data are available, the complete Kalman Filter (KF) conducts the entire filtering process and updates residual and state errors. In the absence of visual data, the partial KF operates only when the time update and filtering state vectors are compensated by the transfer residual, which is estimated through self-propagating state error. The parameters of least-squares support vector machine (LSSVM-NARX) continuously update until the next discontinuous vision observation occurs, with the desired outputs and inner inputs being the error angles and the exogenous inputs being the gyroscope values and time. In the prediction mode, without visual data, LSSVM-NARX can provide estimated angle errors between the complete and partial KFs, which are used to eliminate errors from partial KF output vectors to obtain a reliable and accurate attitude solution. Comparisons and extensive semi-physical simulation experiments under various motion trajectories were performed to validate the effectiveness and superiority of the proposed scheme in an accurate and reliable attitude measurement capability associated with discontinuous observations.