Map-Based Visual-Inertial Monocular SLAM using Inertial assisted Kalman Filter

In this paper, we present a novel tightly-coupled monocular visual-inertial Simultaneous Localization and Mapping algorithm following an inertial assisted Kalman Filter and reusing the estimated 3D map. By leveraging an inertial assisted Kalman Filter, we achieve an efficient motion tracking bearing fast dynamic movement in the front-end. To enable place recognition and reduce the trajectory estimation drift, we construct a factor graph based non-linear optimization in the back-end. We carefully design a feedback mechanism to balance the front/back ends ensuring the estimation accuracy. We also propose a novel initialization method that accurately estimate the scale factor, the gravity, the velocity, and gyroscope and accelerometer biases in a very robust way. We evaluated the algorithm on a public dataset, when compared to other state-of-the-art monocular Visual-Inertial SLAM approaches, our algorithm achieves better accuracy and robustness in an efficient way. By the way, we also evaluate our algorithm in a MonocularInertial setup with a low cost IMU to achieve a robust and lowdrift realtime SLAM system.