Monocular Visual-Inertial Odometry Based on Local Maximum A Posteriori Estimation

This is the authors' abstract. We don't add key points for this paper.

Monocular visual-inertial odometry can effectively solve the problem of unobservable metric scale in monocular odometry. However, due to differences in sensor observation noise and redundant constraints, the positioning accuracy of a general monocular visual-inertial odometer may decline compared with that of the homologous monocular odometer. This letter qualitatively derives the mathematical causes of this phenomenon, introduces the idea of local maximum a posteriori estimation and analyzes its feasibility and generality. Based on this idea, a novel monocular visual-inertial odometry is proposed, including visual tracking, local mapping, and visual-inertial sliding window estimation. Its characteristic is to use the result of the third part combined with appropriate screening strategies to provide necessary prior factors for mapping to complete a factor graph optimization based on local maximum a posteriori estimation, the tracking is used for real-time positioning. This system effectively avoids the problem of decreased tracking accuracy caused by IMU measurements with significant noise. In EuRoC dataset experiments, compared with the homologous visual-inertial odometry system based on maximum a posteriori estimation, the proposed system significantly improves tracking accuracy by 68.6% in position. Compared with other state-of-the-art algorithms, it also shows excellent tracking performance, which proves the superiority of our algorithm.