An Initialization Method for Monocular Visual-Inertial Odometry Based on Uncertainty State

In recent years, monocular visual-inertial systems (VINS) has been widely used and studied in U A V, three-dimensional reconstruction, augmented reality and other fields. Due to the nonlinearity of the whole system, its initialization stage is very critical and vulnerable. The traditional initialization mode is either relatively fixed or has certain requirements for the motion mode. Therefore, in order to overcome this problem and improve the robustness and accuracy of system initialization, we integrate and improve an adaptive monocular visual-inertial SLAM initialization method based on the VINS-Mono[7] framework. Firstly, the initial motion state is judged by the state mode selection algorithm, then two initialization modules are divided to refine the initial mode, realize accurate estimation, and improve the robustness and accuracy of visual inertial joint online initialization. Finally, experiments are carried out on public datasets, and the results show that the proposed method has a greater improvement in convergence and robustness than VINS-Mono.