SLAM System Based on Tightly Coupled Visual-inertial

Abstract

Abstract Aiming at the problem of mobile robot positioning, IMU information is used to improve the performance of visual SLAM. Firstly, instead of the calculation of feature descriptor, LK optical flow is used in the processing of visual information, to increase the speed of feature extraction. And detectors are used to remove outliers. Secondly, the visual information and IMU information are initialized separately to obtain the camera pose and IMU pose. And then Visual-inertia joint initialization is conducted by tightly coupling of the IMU information and the visual information, to improve the calculation. Finally, the pose state estimation is achieved through back-end optimization. The validity of the proposed algorithm is verified by the experiments.