Onboard visual-inertial relative pose and deck motion measurement for autonomous landing

Due to the insufficient accuracy, low frequency, discontinuity and deficiency in deck motion estimation, it is difficult for onboard monocular pose measurement to achieve robust autonomous landing guidance. To address the above issues, an on-board visual-inertial measurement method based on error state Kalman filter is put forward. The proposed framework tightly integrates 2D key points and IMU data to realize efficient and accurate relative pose and deck motion estimation under the constrained condition of dynamic backgrounds, moving target, et al. Considering the motion characteristics of the aircraft and ship, a novel asynchronous error state updating strategy is proposed to achieve high-precision performance. The experimental results demonstrate that the average relative positioning accuracy is improved by about 180% with average translation error decreasing to 3% of the counterpart compared to monocular methods. As to deck motion estimation, the average error of the ship Euler angle is about 0.1°. A cycle of state prediction and update can be conducted within 0.02 ms. The superior performance in accuracy and efficiency of relative and deck motion estimation guarantees significant capacity of the proposed method to integrate with various visual frontends, to perform sound autonomous landing guidance.