A Monocular Visual Inertial Odometer Based on Weighted Median Pre-Integration

In the monocular visual inertial odometer, the graph-based optimization method is more accurate than the filter-based optimization method. However, this performance improvement comes at the cost of generating extensive calculations. So the method of pre-integrating the inertial measurement values between selected keyframes is proposed to reduce the computational effort. The existing pre-integration technology usually uses the numerical integration method of Euler integration and median integration, this method is simple to calculate, but considering the accuracy is lost in the integration process, which has an impact on the accuracy of the final positioning result. Therefore, this paper proposes a weighted median pre-integration method suitable for the monocular visual- inertial odometer, which can effectively improve the positioning accuracy of the monocular visual-inertial odometer. And the effectiveness of the proposed algorithm is proved by experimental results.