Abstract
In the aim of improving the positioning accuracy of the monocular visual-inertial simultaneous localization and mapping (VI-SLAM) system, an improved initialization method with faster convergence is proposed. This approach is classified into three parts: Firstly, in the initial stage, the pure vision measurement model of ORB-SLAM is employed to make all the variables visible. Secondly, the frequency of the IMU and camera was aligned by IMU pre-integration technology. Thirdly, an improved iterative method is put forward for estimating the initial parameters of IMU faster. The estimation of IMU initial parameters is divided into several simpler sub-problems, containing direction refinement gravity estimation, gyroscope deviation estimation, accelerometer bias, and scale estimation. The experimental results on the self-built robot platform show that our method can up-regulate the initialization convergence speed, simultaneously improve the positioning accuracy of the entire VI-SLAM system.
Highlights
Visual simultaneous localization and mapping (VSLAM) techniques allow mobile robots [1,2] and VR/AR devices [3,4] to be aware of their surrounding scene, while carrying on the self-localization in the unknown environments
The monocular camera is not accurate in comparison with a binocular camera, but the computing complexity is lower, the inertial measurement units (IMU) sensor can solve the problem of tracking failure and low precision when the monocular camera moves into the challenging environment by using the IMU preintegration technology and EKF/nonlinear optimization methods
Our initialization method is first integrated into the visual-inertial simultaneous localization and mapping (VI-SLAM) system
Summary
Visual simultaneous localization and mapping (VSLAM) techniques allow mobile robots [1,2] and VR/AR devices [3,4] to be aware of their surrounding scene, while carrying on the self-localization in the unknown environments. The SLAM system based on pure visual sensors has certain problems in robustness and accuracy, which limits its application in the field of terrestrial mobile robots. The accuracy is improved through several visual-inertial bundle adjustments (BA), and the robustness of the system is enhanced with the addition of consensus and observability tests As it is tested on the dataset of Euros [18], it is proved to be consistently initialized with scale errors is less than five percent. The motion of MAV with metric scale can be recovered with a small error, and the accuracy of positioning is maintained at centimeter-level [11] This approach exists several limitations: The process of initial estimation is slow and unstable.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have